Debian Codegraph @

rfc1077.html

raw
1 <pre>Network Working Group Gigabit Working Group
2 Request for Comments: 1077 B. Leiner, Editor
3 November 1988
4
5
6 <span class="h1">Critical Issues in High Bandwidth Networking</span>
7
8
9 Status of this Memo
10
11 This memo presents the results of a working group on High Bandwidth
12 Networking. This RFC is for your information and you are encouraged
13 to comment on the issues presented. Distribution of this memo is
14 unlimited.
15
16 ABSTRACT
17
18 At the request of Maj. Mark Pullen and Maj. Brian Boesch of DARPA, an
19 ad-hoc working group was assembled to develop a set of
20 recommendations on the research required to achieve a ubiquitous
21 high-bandwidth network as discussed in the FCCSET recommendations for
22 Phase III.
23
24 This report outlines a set of research topics aimed at providing the
25 technology base for an interconnected set of networks that can
26 provide highbandwidth capabilities. The suggested research focus
27 draws upon ongoing research and augments it with basic and applied
28 components. The major activities are the development and
29 demonstration of a gigabit backbone network, the development and
30 demonstration of an interconnected set of networks with gigabit
31 throughput and appropriate management techniques, and the development
32 and demonstration of the required overall architecture that allows
33 users to gain access to such high bandwidth.
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52 <span class="grey">Gigabit Working Group [Page 1]</span></pre>
53 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-2" ></span>
54 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
55
56
57 1. Introduction and Summary
58
59
60
61 1.1. Background
62
63
64 The computer communications world is evolving toward both high-
65 bandwidth capability and high-bandwidth requirements. The recent
66 workshop conducted under the auspices of the FCCSET Committee on High
67 Performance Computing [<a href="#ref-1" title="San Diego">1</a>] identified a number of areas where
68 extremely high-bandwidth networking is required to support the
69 scientific research community. These areas range from remote
70 graphical visualization of supercomputer results through the movement
71 of high rate sensor data from space to the ground-based scientific
72 investigator. Similar requirements exist for other applications,
73 such as military command and control (C2) where there is a need to
74 quickly access and act on data obtained from real-time sensors. The
75 workshop identified requirements for switched high-bandwidth service
76 in excess of 300 Mbit/s to a single user, and the need to support
77 service in the range of a Mbit/s on a low-duty-cycle basis to
78 millions of researchers. When added to the needs of the military and
79 commercial users, the aggregate requirement for communications
80 service adds up to many billions of bits per second. The results of
81 this workshop were incorporated into a report by the FCCSET [<a href="#ref-2" title="&quot;A Report to the Congress on Computer Networks to Support Research in the United States: A Study of Critical Problems and Future Options&quot;">2</a>].
82
83 Fortunately, technology is also moving rapidly. Even today, the
84 installed base of fiber optics communications allows us to consider
85 aggregate bandwidths in the range of Gbit/s and beyond to limited
86 geographical regions. Estimates arrived at in the workshop lead one
87 to believe that there will be available raw bandwidth approaching
88 terabits per second.
89
90 The critical question to be addressed is how this raw bandwidth can
91 be used to satisfy the requirements identified in the workshop: 1)
92 provide bandwidth on the order of several Gbit/s to individual users,
93 and 2) provide modest bandwidth on the order of several Mbit/s to a
94 large number of users in a cost-effective manner through the
95 aggregation of their traffic.
96
97 Through its research funding, the Defense Advanced Research Projects
98 Agency (DARPA) has played a central role in the development of
99 packet-oriented communications, which has been of tremendous benefit
100 to the U.S. military in terms of survivability and interoperability.
101 DARPA-funded research has resulted in the ARPANET, the first packet-
102 switched network; the SATNET, MATNET and Wideband Network, which
103 demonstrated the efficient utilization of shared-access satellite
104 channels for communications between geographically diverse sites;
105
106
107
108 <span class="grey">Gigabit Working Group [Page 2]</span></pre>
109 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-3" ></span>
110 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
111
112
113 packet radio networks for mobile tactical environments; the Internet
114 and TCP/IP protocols for interconnection and interoperability between
115 heterogeneous networks and computer systems; the development of
116 electronic mail; and many advances in the areas of network security,
117 privacy, authentication and access control for distributed computing
118 environments. Recognizing DARPA's past accomplishments and its
119 desire to continue to take a leading role in addressing these issues,
120 this document provides a recommendation for research topics in
121 gigabit networking. It is meant to be an organized compendium of the
122 critical research issues to be addressed in developing the technology
123 base needed for such a high bandwidth ubiquitous network.
124
125
126 1.2. Ongoing Activities
127
128
129 The OSTP report referred to above recommended a three-phase approach
130 to achieving the required high-bandwidth networking for the
131 scientific and research community. Some of this work is now well
132 underway. An ad-hoc committee, the Federal Research Internet
133 Coordinating Committee (FRICC) is coordinating the interconnection of
134 the current wide area networking systems in the government; notably
135 those of DARPA, Department of Energy (DoE), National Science
136 Foundation (NSF), National Aeronautics and Space Administration
137 (NASA), and the Department of Health and Human Services (HHS). In
138 accordance with Phases I and II of the OSTP report, this activity
139 will provide for an interconnected set of networks to support
140 research and other scholarly pursuits, and provide a basis for future
141 networking for this community. The networking is being upgraded
142 through shared increased bandwidth (current plans are to share a 45
143 Mbit/s backbone) and coordinated interconnection with the rest of the
144 world. In particular, the FRICC is working with the European
145 networking community under the auspices of another ad-hoc group, the
146 Coordinating Committee for Intercontinental Research Networks
147 (CCIRN), to establish effective US-Europe networking.
148
149 However, as the OSTP recommendations note, the required bandwidth for
150 the future is well beyond currently planned public, private, and
151 government networks. Achieving the required gigabit networking
152 capabilities will require a strong research activity. There is
153 considerable ongoing research in relevant areas that can be drawn
154 upon; particularly in the areas of high-bandwidth communication
155 links, high-speed computer switching, and high-bandwidth local area
156 networks. <a href="#appendix-A">Appendix A</a> provides some pointers to current research
157 efforts.
158
159
160
161
162
163
164 <span class="grey">Gigabit Working Group [Page 3]</span></pre>
165 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-4" ></span>
166 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
167
168
169 1.3. Document Overview
170
171
172 This report outlines a set of research topics aimed at providing the
173 technology base for an interconnected set of networks that can
174 provide the required high-bandwidth capabilities discussed above.
175 The suggested research focus draws upon ongoing research and augments
176 it with basic and applied components. The major activities are the
177 development and demonstration of a Gigabit Backbone network (GB) [<a href="#ref-3" title="the GB">3</a>],
178 the development and demonstration of an interconnected set of
179 networks with gigabit throughput and appropriate management
180 techniques, and the development and demonstration of the required
181 overall architecture that allows users to gain access to such high
182 bandwidth. <a href="#section-2">Section 2</a> discusses functional and performance goals
183 along with the anticipated benefits to the ultimate users of such a
184 system. <a href="#section-3">Section 3</a> provides the discussion of the critical research
185 issues needed to achieve these goals. It is organized into the major
186 areas of technology that need to be addressed: general architectural
187 issues, high-bandwidth switching, high-bandwidth host interfaces,
188 network management algorithms, and network services. The discussion
189 in some cases contains examples of ongoing relevant research or
190 potential approaches. These examples are intended to clarify the
191 issues and not to propose that particular approach. A discussion of
192 the relationship of the suggested research to other ongoing
193 activities and optimal methods for pursuing this research is provided
194 in <a href="#section-4">Section 4</a>.
195
196
197 2. Functional and Performance Goals
198
199
200 In this section, we provide an assessment of the types of services a
201 GN (four or five orders of magnitude faster than the current
202 networks) should provide to its users. In instances where we felt
203 there would be a significant impact on performance, we have provided
204 an estimate of the amount of bandwidth needed and delay allowable to
205 provide these services.
206
207
208 2.1. Networking Application Support
209
210
211 It is envisioned that the GN will be capable of supporting all of the
212 following types of networking applications.
213
214
215
216
217
218
219
220 <span class="grey">Gigabit Working Group [Page 4]</span></pre>
221 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-5" ></span>
222 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
223
224
225 Currently Provided Packet Services
226
227 It is important that the network provide the users with the
228 equivalent of services that are already available in packet-
229 switched networks, such as interactive data exchange, mail
230 service, file transfer, on-line access to remote computing
231 resources, etc., and allow them to expand to other more advanced
232 services to meet their needs as they become available.
233
234 Multi-Media Mail
235
236 This capability will allow users to take advantage of different
237 media types (e.g., graphics, images, voice, and video as well as
238 text and computer data) in the transfer of messages, thereby
239 increasing the effectiveness of message exchange.
240
241 Multi-Media Conferencing
242
243 Such conferencing requires the exchange of large amounts of
244 information in short periods of time. Hence the requirement for
245 high bandwidth at low delay. We estimate that the bandwidth would
246 range from 1.5 to 100 Mbit/s, with an end-to-end delay of no more
247 than a few hundred msec.
248
249 Computer-Generated Real-time Graphics
250
251 Visualizing computer results in the modern world of supercomputers
252 requires large amounts of real time graphics. This in turn will
253 require about 1.5 Mbit/s of bandwidth and no more than several
254 hundred msec. delay.
255
256 High-Speed Transaction Processing
257
258 One of the most important reasons for having an ultra-high-speed
259 network is to take advantage of supercomputing capability. There
260 are several scenarios in which this capability could be utilized.
261 For example, there could be instances where a non-supercomputer
262 may require a supercomputer to perform some processing and provide
263 some intermediate results that will be used to perform still
264 further processing, or the exchange may be between several
265 supercomputers operating in tandem and periodically exchanging
266 results, such as in a battle management, war gaming, or process
267 control applications. In such cases, extremely short response
268 times are necessary to accomplish as many as hundreds of
269 interactions in real time. This requires very high bandwidth, on
270 the order of 100 Mbit/s, and minimum delay, on the order of
271 hundreds of msec.
272
273
274
275
276 <span class="grey">Gigabit Working Group [Page 5]</span></pre>
277 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-6" ></span>
278 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
279
280
281 Wide-Area Distributed Data/Knowledge Base Management Systems
282
283 Computer-stored data, information, and knowledge is distributed
284 around the country for a variety of reasons. The ability to
285 perform complex queries, updates, and report generation as though
286 many large databases are one system would be extremely powerful,
287 yet requires low-delay, high-bandwidth communication for
288 interactive use. The Corporation for National Research
289 Initiatives (NRI) has promoted the notion of a National Knowledge
290 base with these characteristics. In particular, an attractive
291 approach is to cache views at the user sites, or close by to allow
292 efficient repeated queries and multi-relation processing for
293 relations on different nodes. However, with caching, a processing
294 activity may incur a miss in the midst of a query or update,
295 causing it to be delayed by the time required to retrieve the
296 missing relation or portion of relation. To minimize the overhead
297 for cache directories, both at the server and client sites, the
298 unit of caching should be large---say a megabyte or more. In
299 addition, to maintain consistency at the caching client sites,
300 server sites need to multicast invalidations and/or updates.
301 Communication requirements are further increased by replication of
302 the data. The critical parameter is latency for cache misses and
303 consistency operations. Taking the distance between sites to be
304 on average 1/4 the diameter of the country, a one Gbit/s data rate
305 is required to reduce the transmission time to be roughly the same
306 as the propagation delay, namely around 8 milliseconds for this
307 size of unit. Note that this application is supporting far more
308 sophisticated queries and updates than normally associated with
309 transaction processing, thus requiring larger amount of data to be
310 transferred.
311
312
313 2.2. Types of Traffic and Communications Modes
314
315
316 Different types of traffic may impose different constraints in terms
317 of throughput, delay, delay dispersion, reliability and sequenced
318 delivery. Table 1 summarizes some of the main characteristics of
319 several different types of traffic.
320
321
322
323
324
325
326
327
328
329
330
331
332 <span class="grey">Gigabit Working Group [Page 6]</span></pre>
333 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-7" ></span>
334 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
335
336
337 Table 1: Communication Traffic Requirements
338
339 +------------------------+-------------+-------------+-------------+
340 | | | | Error-free |
341 | Traffic | Delay | Throughput | Sequenced |
342 | Type | Requirement | Requirement | Delivery |
343 +------------------------+-------------+-------------+-------------+
344 | Interactive Simulation | Low |Moderate-High| No |
345 +------------------------+-------------+-------------+-------------+
346 | Network Monitoring | Moderate | Low | No |
347 +------------------------+-------------+-------------+-------------+
348 | Virtual Terminal | Low | Low | Yes |
349 +------------------------+-------------+-------------+-------------+
350 | Bulk Transfer | High | High | Yes |
351 +------------------------+-------------+-------------+-------------+
352 | Message | Moderate | Moderate | Yes |
353 +------------------------+-------------+-------------+-------------+
354 | Voice |Low, constant| Moderate | No |
355 +------------------------+-------------+-------------+-------------+
356 | Video |Low, constant| High | No |
357 +------------------------+-------------+-------------+-------------+
358 | Facsimile | Moderate | High | No |
359 +------------------------+-------------+-------------+-------------+
360 | Image Transfer | Variable | High | No |
361 +------------------------+-------------+-------------+-------------+
362 | Distributed Computing | Low | Variable | Yes |
363 +------------------------+-------------+-------------+-------------+
364 | Network Control | Moderate | Low | Yes |
365 +------------------------+-------------+-------------+-------------+
366
367 The topology among users can be of three types: point-to-point (one-
368 to-one connectivity), multicast (one sender and multiple receivers),
369 and conferencing (multiple senders and multiple receivers). There
370 are three types of transfers that can take place among users. They
371 are connection-oriented network service, connectionless network
372 service, and stream or synchronous traffic. Connection and
373 connectionless services are asynchronous. A connection-oriented
374 service assumes and provides for relationships among the multiple
375 packets sent over the connection (e.g., to a common destination)
376 while connectionless service assumes each packet is a complete and
377 separate entity unto itself. For stream or synchronous service a
378 reservation scheme is used to set up and guarantee a constant and
379 steady amount of bandwidth between any two subscribers.
380
381
382
383
384
385
386
387
388 <span class="grey">Gigabit Working Group [Page 7]</span></pre>
389 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-8" ></span>
390 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
391
392
393 2.3. Network Backbone
394
395
396 The GB needs to be of high bandwidth to support a large population of
397 users, and additionally to provide high-speed connectivity among
398 certain subscribers who may need such capability (e.g., between two
399 supercomputers). These users may access the GN from local area
400 networks (LANs) directly connected to the backbone or via high-speed
401 intermediate regional networks. The backbone must also minimize
402 end-to-end delay to support highly interactive high-speed
403 (supercomputer) activities.
404
405 It is important that the LANs that will be connected to the GN be
406 permitted data rates independent of the data rates of the GB. LAN
407 speeds should be allowed to change without affecting the GB, and the
408 GB speeds should be allowed to change without affecting the LANs. In
409 this way, development of the technology for LANs and the GB can
410 proceed independently.
411
412 Access rate requirements to the GB and the GN will vary depending on
413 user requirements and local environments. The users may require
414 access rates ranging from multi-kbit/s in the case of terminals or
415 personal computers connected by modems up to multi-Mbit/s and beyond
416 for powerful workstations up to the Gbit/s range for high-speed
417 computing and data resources.
418
419
420 2.4. Directory Services
421
422
423 Directory services similar to those found in CCITT X.500/ISO DIS 9594
424 need to be provided. These include mapping user names to electronic
425 mail addresses, distribution lists, support for authorization
426 checking, access control, and public key encryption schemes,
427 multimedia mail capabilities, and the ability to keep track of mobile
428 users (those who move from place to place and host computer to host
429 computer). The directory services may also list facilities available
430 to users via the network. Some examples are databases,
431 supercomputing or other special-purpose applications, and on-line
432 help or telephone hotlines.
433
434 The services provided by X.500 may require some extension for GN.
435 For example, there is no provision for multilevel security, and the
436 approach taken to authentication must be studied to ensure that it
437 meets the requirements of GN and its user community.
438
439
440
441
442
443
444 <span class="grey">Gigabit Working Group [Page 8]</span></pre>
445 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-9" ></span>
446 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
447
448
449 2.5. Network Management and Routing
450
451
452 The objective of network management is to ensure that the network
453 functions smoothly and efficiently, and consists of the following:
454 accounting, security, performance monitoring, fault isolation and
455 configuration control.
456
457 Accounting ensures that users are properly billed for the services
458 that the network provides. Accounting enforces a tariff; a tariff
459 expresses a usage policy. The network need only keep track of those
460 items addressed by the tariff, such as allocated bandwidth, number of
461 packets sent, number of ports used, etc. Another type of accounting
462 may need to be supported by the network to support resource sharing,
463 namely accounting analogous to telephone "900" numbers. This
464 accounting performed by the network on behalf of resource providers
465 and consumers is a pragmatic solution to the problem of getting the
466 users and consumers into a financial relationship with each other
467 which has stymied previous attempts to achieve widespread use of
468 specialized resources.
469
470 Performance monitoring is needed so that the managers can tell how
471 the network is performing and take the necessary actions to keep its
472 performance at a level that will provide users with satisfactory
473 service. Fault isolation using technical control mechanisms is
474 needed for network maintenance. Configuration management allows the
475 network to function efficiently.
476
477 Several new types of routing will be required by GN. In addition to
478 true type-of-service, needed to support diverse distributed
479 applications, real-time applications, interactive applications, and
480 bulk data transfer, there will be need for traffic controls to
481 enforce various routing policies. For example, policy may dictate
482 that traffic from certain users, applications, or hosts may not be
483 permitted to traverse certain segments of the network.
484 Alternatively, traffic controls may be used to promote fairness; that
485 is, to make sure that busy link or network segment isn't dominated by
486 a particular source or destination. The ability of applications to
487 reserve network bandwidth in advance of its use, and the use of
488 strategies such as soft connections, will also require development of
489 new routing algorithms.
490
491
492 2.6. Network Security Requirements
493
494
495 Security is a critical factor within the GN and one of those features
496 that are difficult to provide. It is envisioned that both
497
498
499
500 <span class="grey">Gigabit Working Group [Page 9]</span></pre>
501 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-10" ></span>
502 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
503
504
505 unclassified and classified traffic will utilize the GN, so
506 protection mechanisms must be an integral part of the network access
507 strategy. Features such as authentication, integrity,
508 confidentiality, access control, and nonrepudiation are essential to
509 provide trusted and secure communication services for network users.
510
511 A subscriber must have assurance that the person or system he is
512 exchanging information with is indeed who he says he is.
513 Authentication provides this assurance by verifying that the claimed
514 source of a query request, control command, response, etc., is the
515 actual source. Integrity assures that the subscriber's information
516 (such as requests, commands, data, responses, etc.) is not changed,
517 intentionally or unintentionally, while in transit or by replays of
518 earlier traffic. Unauthorized users (e.g., intruders or network
519 viruses) would be denied use of GN assets through access control
520 mechanisms which verify that the authenticated source is authorized
521 to receive the requested information or to initiate the specified
522 command. In addition, nonrepudiation services can be offered to
523 assure a third party that the transmitted information has not been
524 altered. And finally, confidentiality will ensure that the contents
525 of a message are not divulged to unauthorized individuals.
526 Subscribers can decide, based upon their own security needs and
527 particular activities, which of these services are necessary at a
528 given time.
529
530
531 3. Critical Research Issues
532
533
534 In the section above, we discussed the goals of a research program in
535 gigabit networking; namely to provide the technology base for a
536 network that will allow gigabit service to be provided in an
537 effective way. In this section, we discuss those issues which we
538 feel are critical to address in a research program to achieve such
539 goals.
540
541
542 3.1. General Architectural Issues
543
544
545 In the last generation of networks, it was assumed that bandwidth was
546 the scarce resource and the design of the switch was dictated by the
547 need to manage and allocate the bandwidth effectively. The most
548 basic change in the next generation network is that the speeds of the
549 trunks are rising faster than the speeds of the switching elements.
550
551 This change in the balance of speeds has manifested itself in several
552 ways. In most current designs for local area networks, where
553
554
555
556 <span class="grey">Gigabit Working Group [Page 10]</span></pre>
557 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-11" ></span>
558 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
559
560
561 bandwidth is not expensive, the design decision was to trade off
562 effective use of the bandwidth for a simplified switching technique.
563 In particular, networks such as Ethernet use broadcast as the normal
564 distribution method, which essentially eliminates the need for a
565 switching element.
566
567 As we look at still higher speed networks, and in particular networks
568 in which the bandwidth is still the expensive component, we must
569 design new options for switching which will permit effective use of
570 bandwidth without the switch itself becoming the bottleneck.
571
572 The central thrust of new research must thus be to explore new
573 network architectures that are consistent with these very different
574 speed assumptions.
575
576 The development of computer communications has been tremendously
577 distorted by the characteristics of wide-area networking: normally
578 high cost, low speed, high error rate, large delay. The time is ripe
579 for a revolution in thinking, technology, and approaches, analogous
580 to the revolution caused by VCR technology over 8 and 16 mm. film
581 technology.
582
583 Fiber optics is clearly the enabling technology for high-speed
584 transmission, in fact, so much so that there is an expectation that
585 the switching elements will now hold down the data rates. Both
586 conventional circuit switching and packet switching have significant
587 problems at higher data rates. For instance, circuit switching
588 requires increasing delays for FTDM synchronization to handle skew.
589 In the case of packet switching, traditional approaches require too
590 much processing per packet to handle the tremendous data flow. The
591 problem for both switching regimes is the "intelligence" in the
592 switches, which in turn requires electronics technology.
593
594 Besides intelligence, another problem for wide-area networks is
595 storage, both because it ties us to electronics (for the foreseeable
596 future) and because it produces instabilities in a large-scale
597 system. (See, for instance, the work by Van Jacobson on self-
598 organizing phenomena for self-destruction in the Internet.)
599 Techniques are required to eliminate dependence on storage, such as
600 cut-through routing.
601
602 Overall, high-speed WANs are the greatest agents of change, the
603 greatest catalyst both commercially and militarily, and the area ripe
604 for revolution. Judging by the attributes of current high-speed
605 network research prototypes, WANs of the future will be photonic,
606 multi-gigabit networks with enormous throughput, low delay, and low
607 error rate.
608
609
610
611
612 <span class="grey">Gigabit Working Group [Page 11]</span></pre>
613 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-12" ></span>
614 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
615
616
617 A zero-based budgeting approach is required to develop the new high-
618 speed internetwork architecture. That is, the time is ripe to
619 significantly rethink the Internet, building on experience with this
620 system. Issues of concern are manageability, understanding
621 evolvability and support for the new communication requirements,
622 including remote procedure call, real-time, security and fault-
623 tolerance.
624
625 The GN must be able to deal with two sources of high-bandwidth
626 requirements. There will be some end devices (computers) connected
627 more or less directly to the GN because of their individual
628 requirements for high bandwidth (e.g., supercomputers needing to
629 drive remote high-bandwidth graphics devices). In addition, the
630 aggregate traffic due to large numbers of moderate rate users
631 (estimates are roughly up to a million potential users needing up to
632 1 Mbit/s at any given time) results in a high-bandwidth requirement
633 in total on the GN. The statistics of such traffic are different and
634 there are different possible technical approaches for dealing with
635 them. Thus, an architectural approach for dealing with both must be
636 developed.
637
638 Overall, the next-generation architecture has to be, first and
639 foremost, a management architecture. The directions in link speeds,
640 processor speeds and memory solve the performance problems for many
641 communication situations so well that manageability becomes the
642 predominant concern. (In fact, fast communication makes large
643 systems more prone to performance, reliability, and security
644 problems.) In many ways, the management system of the internetwork
645 is the ultimate distributed system. The solution to this tough
646 problem may well require the best talents from the communications,
647 operating systems and distributed systems communities, perhaps even
648 drawing on database and parallelism research.
649
650
651 3.1.1. High-Speed Internet using High-Speed Networks
652
653
654 The GN will need to take advantage of a multitude of different and
655 heterogeneous networks, all of high speed. In addition to networks
656 based on the technology of the GB, there will be high-speed LANs. A
657 key issue in the development of the GN will be the development of a
658 strategy for interconnecting such networks to provide gigabit service
659 on an end to end basis. This will involve techniques for switching,
660 interfacing, and management (as discussed in the sections below)
661 coupled with an architecture that allows the GN to take full
662 advantage of the performance of the various high-speed networks.
663
664
665
666
667
668 <span class="grey">Gigabit Working Group [Page 12]</span></pre>
669 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-13" ></span>
670 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
671
672
673 3.1.2. Network Organization
674
675
676 The GN will need an architecture that supports the need to manage the
677 system as well as obtain high performance. We note that almost all
678 human-engineered systems are hierarchically structured from the
679 standpoint of control, monitoring, and information flow. A
680 hierarchical design may be the key to manageability in the next-
681 generation architecture.
682
683 One approach is to use a general three-level structure, corresponding
684 to interadministrational, intraadministrational, and cluster
685 networks. The first level interconnects communication facilities of
686 truly separate administrations where there is significant separation
687 of security, accounting, and goals. The second level interconnects
688 subadministrations which exist for management convenience in large
689 organizations. For example, a research group within a university may
690 function as a subadministration. The cluster level consists of
691 networks configured to provides maximal performance among hosts which
692 are in frequent communication, such as a set of diskless workstations
693 and their common file server. These hosts are typically, but not
694 necessarily, geographically collocated. For example, two remote
695 networks may be tightly coupled by a fiber optic link that bridges
696 between the two physical networks, making them function as one.
697
698 Research along these lines should study the interorganizational
699 characteristics of communications, such as those being investigated
700 by the IAB Task Force on Autonomous Networks. Based on current
701 results, we expect that such work would clearly demonstrate that
702 considerable communication takes place between particular
703 subadministrations in different administrations; communication
704 patterns are not strictly hierarchical. For example, there might be
705 intense direct communication between the experimental physics
706 departments of two independent universities, or between the computer
707 support group of one company and the operating system development
708 group of another. In addition, (sub)administrations may well also
709 require divisions into public information and private information.
710
711
712 3.1.3. Fault-Tolerant System
713
714
715 Although the GN will be developed as part of an experimental research
716 program, it will also serve as part of the infrastructure for
717 researchers who are experimenting with applications which will use
718 such a network. The GN must have reasonably high availability to
719 support these research activities. In addition to facilitate the
720 transfer of this technology to future operational military and
721
722
723
724 <span class="grey">Gigabit Working Group [Page 13]</span></pre>
725 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-14" ></span>
726 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
727
728
729 commercial users, it will need to be designed to become highly
730 reliable. This can be accomplished through diversity of transmission
731 paths, the development of fault-tolerant switches, use of a
732 distributed control structure with self-correcting algorithms, and
733 the protection of network control traffic. The architecture of a GN
734 should support and allow for all of these things.
735
736
737 3.1.4. Functional Division of Control Between Network Elements
738
739
740 Current protocol architectures use the layered model of functional
741 decomposition first developed in the early work on ARPANET protocols.
742 The concept of layering has been a powerful concept which has allowed
743 dramatic variation in network technologies without requiring the
744 complete reimplementation of applications. The concept of layering
745 has had a first-order impact on the development of international
746 standards for data communication---witness the ISO "Reference Model
747 for Open Systems Interconnection."
748
749 Unfortunately, however, the powerful concept of layering has been
750 paired, both in the DoD Internet work and the ISO work, with an
751 extremely weak concept of the interface between layers. The
752 interface designs are all organized around the idea of commands and
753 responses plus an error indicator. For example, the TCP service
754 interface provides the user with commands to set up or close a TCP
755 connection and commands to send and receive datagrams. The user may
756 well "know" whether they are using a file transfer service or a
757 character-at-a- time virtual terminal, but can't tell the TCP. The
758 underlying network may "know" that failures have reduced the path to
759 the user's destination to a single 9.6 kbit/s link, but it also can't
760 tell the TCP implementation.
761
762 All of the information that an analyst would consider crucial in
763 diagnosing system performance is carefully hidden from adjacent
764 layers. One "solution" often discussed (but rarely implemented) is
765 to condense all of this information into a few bits of "Type of
766 Service" or "Quality of Service" request flowing in one direction
767 only---from application to network. It seems likely that this
768 approach cannot succeed, both because it applies too much compression
769 to the knowledge available and because it does not provide two-way
770 flow.
771
772 We believe it to be likely that the next-generation network will
773 require a much richer interface between every pair of adjacent layers
774 if adequate performance is to be achieved. Research is needed into
775 the conceptual mechanisms, both indicators and controls, that can be
776 implemented at these interfaces and that, when used, will result in
777
778
779
780 <span class="grey">Gigabit Working Group [Page 14]</span></pre>
781 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-15" ></span>
782 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
783
784
785 better performance. If real differences in performance can be
786 observed, then the implementors of every layer will have a strong
787 incentive to make use of the mechanisms.
788
789 We can observe the first glimmers of this sort of coordination
790 between layers in current work. For example, in the ISO work there
791 are 5 classes of transport protocol which are supposed to provide a
792 range of possible matches between application needs and network
793 capabilities. Unfortunately, it is the case today that the class of
794 transport protocol is chosen statically, by the implementer, rather
795 than dynamically. The DARPA Wideband net offers a choice of stream
796 or datagram service, but typically a given host uses all one or all
797 the other---again, a static rather than a dynamic choice. The
798 research that we believe is needed, therefore, is not how to provide
799 alternatives, but how to provide them and choose among them on a
800 dynamic, real-time basis.
801
802
803 3.1.5. Different Switch Technologies
804
805
806 One approach to high-performance networking is to design a technology
807 that is expected to work as a stand-alone demonstration, without
808 addressing the need for interconnection to other networks. Such an
809 experiment may be very valuable for rapid exploration of the design
810 space. However, our experience with the Internet project suggests
811 that a primary research goal should be the development of a network
812 architecture that permits the interconnection of a number of
813 different switching technologies.
814
815 The Internet project was successful to a large extent because it
816 could incorporate a number of new and preexisting network
817 technologies: various local area networks, store and forward
818 switching networks, broadcast satellite nets, packet radio networks,
819 and so on. In this way, it decoupled the use of the protocols from a
820 particular technology base. In fact, the technology base evolved
821 rapidly, but the Internet protocols themselves provided a stability
822 that led to their success.
823
824 The next-generation architecture must similarly deal with a diverse
825 and evolving technology base. We see "fast-packet" switching now
826 being developed (for example in B-ISDN); we see photonic switching
827 and wavelength division multiplexing as more advanced technologies.
828 We must divorce our architecture from dependence on any one of these.
829
830 At the host interface, we must divorce the multiplexing of the medium
831 from the form of data that the host sees. Today the packet is used
832 both as multiplexing and interface element. In the future, the host
833
834
835
836 <span class="grey">Gigabit Working Group [Page 15]</span></pre>
837 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-16" ></span>
838 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
839
840
841 may see the network as a message-passing system, or as memory. At
842 the same time, the network may use classic packets, wavelength
843 division, or space division switching.
844
845 A number of basic functions must be rethought to provide an
846 architecture that is not dependent on the underlying switching model.
847 For example, our transport protocols assume that data will be lost in
848 units of a packet. If part of a packet is lost, we discard the whole
849 thing. And if several packets are systematically lost in sequence,
850 we may not recover effectively. There must be a host-level unit of
851 error recovery that is independent of the network. This sort of
852 abstraction must be applied to all the aspects of service
853 specification: error recovery, flow control, addressing, and so on.
854
855
856 3.1.6. Network Operations, Monitoring, and Control
857
858
859 There is a hierarchy of progressively more effective and
860 sophisticated techniques for network management that applies
861 regardless of network bandwidth and application considerations:
862
863 1. Reactive problem management
864
865 2. Reactive resource management
866
867 3. Proactive problem management
868
869 4. Proactive resource management.
870
871 Today's network management strategies are primarily reactive rather
872 than proactive: Problem management is initiated in response to user
873 complaints about service outages; resource allocation decisions are
874 made when users complain about deterioration of quality of service.
875 Today's network management systems are stuck at step 1 or perhaps
876 step 2 of the hierarchy.
877
878 Future network management systems will provide proactive problem
879 management---problem diagnosis and restoral of service before users
880 become aware that there was a problem; and proactive resource
881 management---dynamic allocation of network bandwidth and switching
882 resources to ensure that an acceptable level of service is
883 continuously maintained.
884
885 The GN management system should be expected to provide proactive
886 problem and resource management capabilities. It will have to do so
887 while contending with three important changes in the managed network
888 environment:
889
890
891
892 <span class="grey">Gigabit Working Group [Page 16]</span></pre>
893 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-17" ></span>
894 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
895
896
897 1. More complicated devices under management
898
899 2. More diverse types of devices
900
901 3. More variety of application protocols.
902
903 Performance under these conditions will require that we seriously
904 re-think how a network management system handles the expected high
905 volumes of raw management-related data. It will become especially
906 important for the system to provide thresholding, filtering, and
907 alerting mechanisms that can save the human operator from drowning in
908 data, while still permitting access to details when diagnostic or
909 fault isolation modes are invoked.
910
911 The presence of expert assistant capabilities for early fault
912 detection, diagnosis, and problem resolution will be mandatory.
913 These capabilities are highly desirable today, but they will be
914 essential to contend with the complexity and diversity of devices and
915 applications in the Gigabit Network.
916
917 In addition to its role in dealing with complexity, automation
918 provides the only hope of controlling and reducing the high costs of
919 daily management and operation of a GN.
920
921 Proactive resource management in GNs must be better understood and
922 practiced, initially as an effort requiring human intervention and
923 direction. Once this is achieved, it too must become automated to a
924 high degree in the GN.
925
926
927 3.1.7. Naming and Addressing Strategies
928
929
930 Current networks, both voice (telephone) and data, use addressing
931 structures which closely tie the address to the physical location on
932 the network. That is, the address identifies a physical access
933 point, rather than the higher-level entity (computer, process, human)
934 attached to that access point. In future networks, this physical
935 aspect of addressing must be removed.
936
937 Consider, for example, finding the desired party in the telephone
938 network of today. For a person not at his listed number, finding the
939 number of the correct telephone may require preliminary calls, in
940 which advice is given to the person placing the call. This works
941 well when a human is placing the call, since humans are well equipped
942 to cope with arbitrary conversations. But if a computer is placing
943 the call, the process of obtaining the correct address will have to
944 be incorporated in the architecture as a core service of the network.
945
946
947
948 <span class="grey">Gigabit Working Group [Page 17]</span></pre>
949 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-18" ></span>
950 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
951
952
953 Since it is reasonable to expect mobile hosts, hosts that are
954 connected to multiple networks, and replicated hosts, the issue of
955 mapping to the physical address must be properly resolved.
956
957 To permit the network to maintain the dynamic mapping to current
958 physical address, it is necessary that high-level entities have a
959 name (or logical address) that identifies them independently of
960 location. The name is maintained by the network, and mapped to the
961 current physical location as a core network service. For example,
962 mobile hosts, hosts that are connected to multiple networks, and
963 replicated hosts would have static names whose mapping to physical
964 addresses (many-to-one, in some cases) would change with time.
965
966 Hosts are not the only entities whose physical location varies.
967 Users' electronic mail addresses change. Within distributed systems,
968 processes and files migrate from host to host. In a computing
969 environment where robustness and survivability are important, entire
970 applications may move about, or they may be redundant.
971
972 The needed function must be considered in the context of the mobility
973 and address resolution rates if all addresses in a global data
974 network were of this sort. The distributed network directory
975 discussed elsewhere in this report should be designed to provide the
976 necessary flexibility, and responsiveness. The nature and
977 administration of names must also be considered.
978
979 Names that are arbitrary or unwieldy would be barely better than the
980 addresses used now. The name space should be designed so that it can
981 easily be partitioned among the agencies that will assign names. The
982 structure of names should facilitate, rather than hinder, the mapping
983 function. For example, it would be hard to optimize the mapping
984 function if names were flat and unstructured.
985
986
987 3.2. High-Speed Switching
988
989
990 The term "high-speed switching" refers to changing the switching at a
991 high rate, rather than switching high-speed links, because the latter
992 is not difficult at low speeds. (Consider, for example, manual
993 switching of fiber connections). The switching regime chosen for the
994 network determines various aspects of its performance, its charging
995 policies, and even its effective capabilities. As an example of the
996 latter, it is difficult to expect a circuit-switched network to
997 provide strong multicast support.
998
999 A major area of debate lies in the choice between packet switching
1000 and circuit switching. This is a key research issue for the GN,
1001
1002
1003
1004 <span class="grey">Gigabit Working Group [Page 18]</span></pre>
1005 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-19" ></span>
1006 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1007
1008
1009 considering also the possibility of there being combinations of the
1010 two approaches that are feasible.
1011
1012
1013 3.2.1. Unit of Management vs. Multiplexing
1014
1015
1016 With very high data rates, either the unit of management and
1017 switching must be larger or the speed of the processor elements for
1018 management and switching must be faster. For example, at a gigabit,
1019 a 576 byte packet takes roughly 5 microseconds to be received so a
1020 packet switch must act extremely fast to avoid being the dominant
1021 delay in packet times. Moreover, the storage time for the packet in
1022 a conventional store and forward implementation also becomes a
1023 significant component of the delay. Thus, for packet switching to
1024 remain attractive in this environment, it appears necessary to
1025 increase the size of packets (or switch on packet groups), do so-
1026 called virtual cut-through and use high-speed routing techniques,
1027 such as high-speed route caches and source routing.
1028
1029 Alternatively, for circuit switching to be attractive, it must
1030 provide very fast circuit setup and tear-down to support the bursty
1031 nature of most computer communication. This problem is rendered
1032 difficult (and perhaps impossible for certain traffic loads) because
1033 the delay across the country is so large relative to the data rate.
1034 That is, even with techniques such as so-called fast select,
1035 bandwidth is reserved by the circuit along the path for almost twice
1036 the propagation time before being used.
1037
1038 With gigabit circuit switching, because it is not feasible to
1039 physically switch channels, the low-level switching is likely doing
1040 FTDM on micro-packets, as is currently done in telephony. Performing
1041 FTDM at gigabit data rates is a challenging research problem if the
1042 skew introduced by wide-area communication is to be handled with
1043 reasonable overhead for spacing of this micro-packets. Given the
1044 lead and resources of the telephone companies, this area of
1045 investigation should, if pursued, be pursued cooperatively.
1046
1047
1048 3.2.2. Bandwidth Reservation Algorithms
1049
1050
1051 Some applications, such as real-time video, require sustained high
1052 data rate streams over a significant period of time, such as minutes
1053 if not hours. Intuitively, it is appealing for such applications to
1054 pre-allocate the bandwidth they require to minimize the switching
1055 load on the network and guarantee that the required bandwidth is
1056 available. Research is required to determine the merits of bandwidth
1057
1058
1059
1060 <span class="grey">Gigabit Working Group [Page 19]</span></pre>
1061 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-20" ></span>
1062 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1063
1064
1065 reservation, particular in conjunction with the different switching
1066 technologies. There is some concern to raise that bandwidth
1067 reservation may require excessive intelligence in the network,
1068 reducing the performance and reliability of the network. In
1069 addition, bandwidth reservation opens a new option for denial of
1070 service by an intruder or malicious user. Thus, investigations in
1071 this area need to proceed in concert with work on switching
1072 technologies and capabilities and security and reliability
1073 requirements.
1074
1075
1076 3.2.3. Multicast Capabilities
1077
1078
1079 It is now widely accepted that multicast should be provided as a
1080 user-level service, as described in <a href="./rfc1054">RFC 1054</a> for IP, for example.
1081 However, further research is required to determine the best way to
1082 support this facility at the network layer and lower. It is fairly
1083 clear that the GN will be built from point-to-point fiber links that
1084 do not provide multicast/broadcast for free. At the most
1085 conservative extreme, one could provide no support and require that
1086 each host or gateway simulate multicast by sending multiple,
1087 individually addressed packets. However, there are significant
1088 advantages to providing very low level multicast support (besides the
1089 obvious performance advantages). For example, multicast routing in a
1090 flooding form provides the most fault-tolerant, lowest-delay form of
1091 delivery which, if reserved for very high priority messages, provides
1092 a good emergency facility for high-stress network applications.
1093 Multicast may also be useful as an approach to defeat traffic
1094 analysis.
1095
1096 Another key issue arises with the distinction between so-called open
1097 group multicast and closed group multicast. In the former, any host
1098 can multicast to the group, whereas in the latter, only members of
1099 the group can multicast to it. The latter is easier to support and
1100 adequate for conferencing, for example. However, for more client-
1101 server structured applications, such as using file/database server,
1102 computation servers, etc. as groups, open multicast is required.
1103 Research is needed to address both forms of multicast. In addition,
1104 security issues arise in controlling the membership of multicast
1105 groups. This issue should be addressed in concert with work on
1106 secure forms of routing in general.
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116 <span class="grey">Gigabit Working Group [Page 20]</span></pre>
1117 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-21" ></span>
1118 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1119
1120
1121 3.2.4. Gateway Technologies
1122
1123
1124 With the wide-area interconnection of local networks by the GN,
1125 gateways are expected to become a significant performance bottleneck
1126 unless significant advances are made in gateway performance. In
1127 addition, many network management concerns suggest putting more
1128 functionality (such as access control) in the gateways, further
1129 increasing their load and the need for greater capacity. This would
1130 then raise the issue of the trade-off between general-purpose
1131 hardware and special-purpose hardware.
1132
1133 On the general-purpose side, it may be feasible to use a general-
1134 purpose multiprocessor based on high-end microprocessors (perhaps as
1135 exotic as the GaAs MIPS) in conjunction with a high-speed block
1136 transfer bus, as proposed as part of the FutureBus standard (which is
1137 extendible to higher speeds than currently commercially planned) and
1138 intelligent high-speed network adaptors. This would also allow the
1139 direct use of hardware, operating systems, and software tools
1140 developed as part of other DARPA programs, such as Strategic
1141 Computing. It also appears to make this gateway software more
1142 portable to commercial machines as they become available in this
1143 performance range.
1144
1145 The specialized hardware approach is based on the assumption that
1146 general-purpose hardware, particularly the interconnection bus,
1147 cannot be fast enough to support the level of performance required.
1148 The expected emphasis is on various interconnection network
1149 techniques. These approaches appear to require greater expense, less
1150 commercial availability and more specialized software. They need to
1151 be critically evaluated with respect to the general-purpose gateway
1152 hardware approach, especially if the latter is using multiple buses
1153 for fault-tolerance as well as capacity extension (in the absence of
1154 failure).
1155
1156 The same general-purpose vs. special-purpose contention is an issue
1157 with operating system software. Conventionally, gateways run
1158 specialized run-time executives that are designed specifically for
1159 the gateway and gateway functions. However, the growing
1160 sophistication of the gateway makes this approach less feasible. It
1161 appears important to investigate the feasibility of using a standard
1162 operating system foundation on the gateways that is known to provide
1163 the required security and reliability properties (as well as real-
1164 time performance properties).
1165
1166
1167
1168
1169
1170
1171
1172 <span class="grey">Gigabit Working Group [Page 21]</span></pre>
1173 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-22" ></span>
1174 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1175
1176
1177 3.2.5. VLSI and Optronics Implementations
1178
1179
1180 It appears fairly clear that gigabit communication will use fiber
1181 optics for at least the near future. Without major advances in
1182 optronics to allow effectively for optical computers, communication
1183 must cross the optical-electronic boundary two or more times. There
1184 are significant cost, performance, reliability, and security benefits
1185 for minimizing the number of such crossings. (As an example of a
1186 security benefit, optics is not prone to electronic surveillance or
1187 jamming while electronics clearly is, so replacing an optic-
1188 electronic-optic node with a pure optic node eliminates that
1189 vulnerability point.)
1190
1191 The benefits of improved technology in optronics is so great that its
1192 application here is purely another motivation for an already active
1193 research area (that deserves strong continued support). Therefore,
1194 we focus here in the issue of matching current (and near-term
1195 expected) optronics capabilities with network requirements.
1196
1197 The first and perhaps greatest area of opportunity is to achieve
1198 totally (or largely) photonic switches in the network switching
1199 nodes. That is, most packets would be switched without crossing the
1200 optics-electronics boundary at all. For this to be feasible, the
1201 switch must use very simple switching logic, require very little
1202 storage and operate on packets of a significant size. The source-
1203 routed packet switches with loopback on blockage of Blazenet
1204 illustrate the type of techniques that appear required to achieve
1205 this goal.
1206
1207 Research is required to investigate the feasibility of optronic
1208 implementation of switches. It appears highly likely that networks
1209 will at some point in the future be totally photonically switched,
1210 having the impact on networking comparable to the effect of
1211 integrated circuits on processors and memories.
1212
1213 A next level of focus is to achieve optical switching in the common
1214 case in gateways. One model is a multiprocessor with an optical
1215 interconnect. Packets associated with established paths through the
1216 gateway are optically switched and processed through the
1217 interconnect. Other packets are routed to the multiprocessor,
1218 crossing into the electronics domain. Research is required to marry
1219 the networking requirements and technology with optronics technology,
1220 pushing the state of the art in both areas in the process.
1221
1222 Given the long-term presence of the optic-electronic boundary,
1223 improvements in technology in this area are also important. However,
1224 it appears that there is already enormous commercial research
1225
1226
1227
1228 <span class="grey">Gigabit Working Group [Page 22]</span></pre>
1229 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-23" ></span>
1230 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1231
1232
1233 activity in this area, particularly within the telephone companies.
1234 This is another area in which collaborative investigation appears far
1235 better than an new independent research effort.
1236
1237 VLSI technology is an established technology with active research
1238 support. The GN effort does not appear to require major new
1239 initiatives in the VLSI area, yet one should be open to significant
1240 novel opportunities not identified here.
1241
1242
1243 3.2.6. High-Speed Transfer Protocols
1244
1245
1246 To achieve the desired speeds, it will be necessary to rethink the
1247 form of protocols.
1248
1249 1. The simple idea of a stateless gateway must be replaced by a
1250 more complex model in which the gateway understands the
1251 desired function of the end point and applies suitable
1252 optimizations to the flow.
1253
1254 2. If multiplexing is done in the time domain, the elements of
1255 multiplexing are probably so small that no significant
1256 processing can be performed on each individually. They must
1257 be processed as an aggregate. This implies that the unit of
1258 multiplexing is not the same as the unit of processing.
1259
1260 3. The interfaces between the structural layers of the
1261 communication system must change from a simple
1262 command/response style to a richer system which includes
1263 indications and controls.
1264
1265 4. An approach must be developed that couples the memory
1266 management in the host and the structure of the transmitted
1267 data, to allow efficient transfers into host memory.
1268
1269 The result of rethinking these problems will be a new style of
1270 communications and protocols, in which there is a much higher degree
1271 of shared responsibility among the components (hosts, switches,
1272 gateways). This may have little resemblance to previous work either
1273 in the DARPA or commercial communities.
1274
1275
1276 3.3. High-Speed Host Interfaces
1277
1278
1279 As networks get faster, the most significant bottleneck will turn out
1280 to be the packet processing overhead in the host. While this does
1281
1282
1283
1284 <span class="grey">Gigabit Working Group [Page 23]</span></pre>
1285 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-24" ></span>
1286 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1287
1288
1289 not restrict the aggregate rates we can achieve over trunks, it
1290 prevents delivery of high data rate flows to the host-based
1291 applications, which will prevent the development of new applications
1292 needing high bandwidth. The host bottleneck is thus a serious
1293 impediment to networked use of supercomputers.
1294
1295 To build a GN we need to create new ways for hosts and their high
1296 bandwidth peripherals to connect to networks. We believe that
1297 pursuing research in the ways to most effectively isolate host and
1298 LAN development paths from the GN is the most productive way to
1299 proceed. By decoupling the development paths, neither is restricted
1300 by the momentary performance of capability bottlenecks of the other.
1301 The best context in which to view this separation is with the notion
1302 of a network front end (NFE). The NFE can take the electronic input
1303 data at many data rates and transform it into gigabit light data
1304 appropriately packetized to traverse the GN. The NFE can accept
1305 inputs from many types of gateways, hosts, host peripherals, and LANS
1306 and provide arbitration and path set-up facilities as needed. Most
1307 importantly, the NFE can perform protocol arbitration to retain
1308 upward compatibility with the existing Internet protocols while
1309 enabling those sophisticated network input sources to execute GN
1310 specific high-throughput protocols. Of course, this introduces the
1311 need for research into high-speed NFEs to avoid the NFE becoming a
1312 bottleneck.
1313
1314
1315 3.3.1. VLSI and Optronics Implementations
1316
1317
1318 In a host interface, unless the host is optical (an unlikely prospect
1319 in the near-term), the opportunities for optronic support are
1320 limited. In fact, with a serial-to-parallel conversion on reception
1321 stepping the clock rate down by a factor of 32 (assuming a 32-bit
1322 data path on the host interface), optronic speeds are not required in
1323 the immediate future.
1324
1325 One exception may be for encryption. Current VLSI implementations of
1326 standard encryption algorithms run in the 10 Mbit/s range. Optronic
1327 implementation of these encryption techniques and encryption
1328 techniques specifically oriented to, or taking advantage of, optronic
1329 capabilities appears to be an area of some potential (and enormous
1330 benefit if achieved).
1331
1332 The potential of targeted VLSI research in this area appears limited
1333 for similar reasons discussed above with its application in high-
1334 speed switching. The major benefits will arise from work that is
1335 well-motivated by other research (such as high-performance
1336 parallelism) and by strong commercial interest. Again, we need to be
1337
1338
1339
1340 <span class="grey">Gigabit Working Group [Page 24]</span></pre>
1341 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-25" ></span>
1342 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1343
1344
1345 open to imaginative opportunities not foreseen here while keeping
1346 ourselves from being diverted into low-impact research without
1347 further insights being put forward.
1348
1349
1350 3.3.2. High-Performance Transport Protocols
1351
1352
1353 Current transport protocols exhibit some severe problems for maximal
1354 performance, especially for using hardware support. For example, TCP
1355 places the checksum in the packet header, forcing the packet to be
1356 formed and read fully before transmission begins. ISO TP4 is even
1357 worse, locating the checksum in a variable portion of the header at
1358 an indeterminate offset, making hardware implementation extremely
1359 difficult.
1360
1361 The current Internet has thrived and grown due to the existence of
1362 TCP implementations for a wide variety of classes of host computers.
1363 These various TCP implementations achieve robust interoperability by
1364 a "least common denominator" approach to features and options. Some
1365 applications have arisen in the current Internet, and analogs can be
1366 envisioned for the GN environment, which need qualities of service
1367 not generally supported by the ubiquitous generic TCP, and therefore
1368 special purpose transport protocols have been developed. Examples
1369 include special purpose transport protocols such as UDP (user
1370 datagram protocol), RDP (reliable datagram protocol), LDP
1371 (loader/debugger protocol), NETBLT (high-speed block transfer
1372 protocol), NVP (network voice protocol) and PVP (packet video
1373 protocol). Efforts are also under way to develop a new generic
1374 transport protocol VMTP (versatile message transaction protocol)
1375 which will remedy some of deficiencies of TCP, without the need to
1376 resort to special purpose protocols for some applications. Research
1377 is needed in this area to understand how transport level protocols
1378 should be constructed for a GN which provide adequate qualities of
1379 service and ease of implementation.
1380
1381 A new transport protocol of reasonable success can be expected to
1382 last for ten years more. Therefore, a new protocol should not be
1383 over optimized for current networks and must not ignore the
1384 functional deficiencies of current protocols. These deficiencies are
1385 essential to remedy before it is feasible to deploy even current
1386 distributed systems technology for military and commercial
1387 applications.
1388
1389 Forward Error Correction (FEC) is a useful approach when the
1390 bandwidth/delay ratio of the physical medium is high, as can be
1391 expected in transcontinental photonic links. A degenerate form of
1392 FEC is to simply transmit multiple copies of the data; this allows
1393
1394
1395
1396 <span class="grey">Gigabit Working Group [Page 25]</span></pre>
1397 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-26" ></span>
1398 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1399
1400
1401 one to trade bandwidth for delay and reliability, without requiring
1402 much intelligence. In fact, it is generally true that reliability,
1403 bandwidth, and delay are interrelated and an improvement in one
1404 generally comes at the expense of the others for a given technology.
1405 Research is required to find appropriate operating points in networks
1406 using transmission components which offer extremely high bandwidth
1407 with very good bit-error-rate performance.
1408
1409
1410 3.3.3. Network Adaptors
1411
1412
1413 With the promised speed of networks, the future network adaptor must
1414 be viewed as a memory interconnect, tying the memory in one host to
1415 another, at least if the data rate and the low latency made possible
1416 by the network is to be realized at the host-to-host or process-to-
1417 process level. The challenge is too great to be met by just
1418 implementing protocols in custom VLSI.
1419
1420 Research is required to investigate the impact of network
1421 interconnection on a machine architecture and to define and evaluate
1422 new network adaptor architectures. Of key importance is integration
1423 of network adaptor into the operating system so that process-to-
1424 process communications performance matches that offered by the
1425 network. In particular, we conjecture that the transport level will
1426 be implemented largely, if not entirely, in the network adaptor,
1427 providing the host with reliable memory-to-memory transfer at memory
1428 speeds with a minimum of interrupt processing bus overhead and packet
1429 processing.
1430
1431 Drawing an analogy to RISC technology again, maximal performance
1432 requires a well-designed and coordinated protocol, software, and
1433 hardware (network adaptor) design. Current standard protocols are
1434 significantly flawed for hardware compatibility, suggesting a need
1435 for considerable further research on high-performance protocol
1436 design.
1437
1438
1439 3.3.4. Host Operating System Software
1440
1441
1442 Conventionally, communication has been an add-on to an operating
1443 system. With the GN, the network may well become the fastest
1444 "peripheral" connected to most nodes. High-performance process-to-
1445 process (or application to application) communication will not be
1446 achieved until the operating system is well designed for fast access
1447 to and from the network. For example, incorporating templates of the
1448 network packet header directly in the process descriptor may allow a
1449
1450
1451
1452 <span class="grey">Gigabit Working Group [Page 26]</span></pre>
1453 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-27" ></span>
1454 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1455
1456
1457 process to initiate communications with minimal overhead. Similarly,
1458 memory mapping can be used to eliminate copies between data arriving
1459 from the network and it being delivered to the applications. With a
1460 GN, an extra copy forced by the operating system may easily double
1461 the perceived transfer time for a packet between applications.
1462
1463 Besides matching data transfer mechanisms, operating systems must be
1464 well-matched in security design to that supported by the host
1465 interface and network as well. Otherwise, all but the most trivial
1466 additional security actions by the operating system in common case
1467 communication can easily eliminate the performance benefits of the
1468 GN. For example, if the host has to do further encryption or
1469 decryption, the throughput is likely to be at least halved and the
1470 latency doubled.
1471
1472 Research effort is required to further refine operating systems for
1473 the level of performance offered by the GN. This effort may well be
1474 best realized with coupling existing efforts in distributed systems
1475 with the GN activities, as opposed to starting new separate efforts.
1476
1477
1478 3.4. Advanced Network Management Algorithms
1479
1480
1481 An important emphasis for research into network management should be
1482 on decentralized approaches. The ratio of propagation delay across
1483 the country to data rates in a GN appear to be too great to deal
1484 effectively with resource management centrally when traffic load is
1485 bursty and unstable (and if it is not, one might argue there is no
1486 problem). In addition, important principles of fault containment and
1487 minimal privilege for reliability and security suggest that a
1488 centralized management approach is infeasible. In particular,
1489 compromising the security of one portion of the network should not
1490 compromise the security of the whole network. Similarly, a failure
1491 or fault should affect at most a local region of the network.
1492
1493 The challenge is clearly to provide decentralized management
1494 techniques that lead to good global behavior in the normal case and
1495 acceptable behavior in expected worst-case failures, traffic
1496 variations and security intrusions.
1497
1498
1499 3.4.1. Control Flow vs. Data Flow
1500
1501
1502 Network operational communications can be separated into flow of user
1503 data and flow of management/control data. However, the user data
1504 must contain some amount of control data. One question that needs to
1505
1506
1507
1508 <span class="grey">Gigabit Working Group [Page 27]</span></pre>
1509 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-28" ></span>
1510 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1511
1512
1513 be explored in light of changes in communications and computing costs
1514 and performance is the trade-off between these two flows. An example
1515 of a potential approach is to use data units which contain predefined
1516 path indicators. The switch can perform a simple table look-up which
1517 maps the path indicator onto the preferred outbound link and
1518 transmits the packet immediately. There is a path set-up packet
1519 which fills in the appropriate tables. Path set-up occurs before the
1520 first data packet flows and then, while data is flowing, to improve
1521 the routes during the lifetime of the connection. This concept has
1522 been discussed in the Internet engineering group under the name of
1523 soft connections.
1524
1525 We note that separating the data flow from the control flow in the GN
1526 has security and reliability advantages as well. We could encrypt
1527 most of the packet header to provide confidentiality within the GN
1528 and to limit the ability of intruders to perform traffic analysis.
1529 And, by separating the control flow, we can encrypt all the control
1530 exchanges between switches and the host front ends thereby offering
1531 confidentiality and integrity. No unauthorized entity will be able
1532 to alter or examine the control traffic. By employing a path set-up
1533 procedure, we can assure that the GN NFE-to-NFE path is functioning
1534 and also include user-specific requirements in the route. For
1535 example, we could request a certain bandwidth allocation and simplify
1536 the job of the switches in handling flow control. We could also set
1537 up backup paths in case the output link will be busy for so many
1538 microseconds that the packet cannot be stored until the link is
1539 freed.
1540
1541
1542 3.4.2. Resource Management Algorithms
1543
1544
1545 Most current networks deliver one quality of service. X.25 networks
1546 deliver a reliable byte-stream. Most LANs deliver a best-effort
1547 unreliable service. There are few networks today that can support
1548 multiple types of service, and allocate their resources among them.
1549 Indeed, for many networks, such as best-effort unreliable service,
1550 there is little management of resources at all. The next generation
1551 of network will require a much more controlled allocation of
1552 resources.
1553
1554 There will be a much wider range of desired types of service, with
1555 current services such as remote procedure call mixing with new
1556 services such as video streams. Unless these are separately
1557 recognized and controlled, there is little reason to believe that
1558 effective service can be delivered unless the network is very lightly
1559 loaded.
1560
1561
1562
1563
1564 <span class="grey">Gigabit Working Group [Page 28]</span></pre>
1565 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-29" ></span>
1566 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1567
1568
1569 In order to support multiple types of service, two things must
1570 happen, both a change from current practice. First, the application
1571 must describe to the network what type of service is required.
1572 Second, the network must use this information to make resource
1573 allocation decisions. Both of these practices present difficulties.
1574
1575 Past experience suggests that application code is not prepared to
1576 know or specify what service it needs. By custom, operating systems
1577 provide a virtual world, and the applications in this world are
1578 unaware of the relation between this and the reality of time and
1579 space. Resource requests must be in real terms. Allocation of
1580 resources in the network is difficult, because it requires that
1581 decisions be made in the network, but as network packet throughput
1582 increases, there is less time for decisions.
1583
1584 The resolution of this latter conflict is to observe that decisions
1585 must be made on larger units than the unit of multiplexing such as
1586 the packet. This in turn implies that packets must be visible to the
1587 network as being part of a sequence, as opposed to the pure datagram
1588 model previously exploited. As suggested earlier in this report,
1589 research is required to support this more complex form of switch
1590 without compromising robustness.
1591
1592 To permit the application to specify the service it needs, it will be
1593 necessary to propose some abstraction of service class. By clever
1594 design of this abstraction, it should be possible to allow the
1595 application to describe its needs effectively. For example, an
1596 application such as file transfer or mail has two modes of operation;
1597 bulk data transfer and remote procedure call. The application may
1598 not be able to predict when it will be in which mode, but if it just
1599 describes both of them, the system may be able to adapt by observing
1600 its current operation.
1601
1602 Experimentation needs to be done to determine a suitable service
1603 specification interface. This experimentation could be done in the
1604 context of the current protocols, and could thus be undertaken at
1605 once.
1606
1607
1608 3.4.3. Adaptive Protocols
1609
1610
1611 Network operating conditions can vary quickly and over a wide range.
1612 This is true of the current Internet, and is likely to affect the GN
1613 too. Protocols that can adapt to changing circumstances would
1614 provide more even and robust service than is currently possible. For
1615 example, when error rates increased, a protocol implementation might
1616 decide to use smaller packets, thus reducing the burden caused by
1617
1618
1619
1620 <span class="grey">Gigabit Working Group [Page 29]</span></pre>
1621 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-30" ></span>
1622 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1623
1624
1625 retransmissions.
1626
1627 The environment in which a protocol operates can be described in
1628 terms of the service it is getting from the next lower layer. A
1629 protocol implementation can adapt to changes in that service by
1630 tuning its internal mechanisms (time-outs, retransmission strategies,
1631 etc.). Therefore, to design adaptive protocols, we must understand
1632 the interaction between protocol layers and the mechanisms used
1633 within them. There has been some work done in this area. For
1634 example, the SATNET measurement task force has looked at the
1635 interactions between the protocol used by the SIMP, IP, and TCP.
1636 What is needed is a more complete characterization of the
1637 interactions at various layer boundaries, and the development of
1638 appropriate protocol designs and mechanisms to provide for necessary
1639 adaptations and renegotiations.
1640
1641
1642 3.4.4. Error Recovery Mechanisms
1643
1644
1645 Being large and complex, GNs will experience a variety of faults such
1646 as link or nodal failure, excessive buffer overflow due to faulty
1647 flow and congestion control, and partial failure of switching fabric.
1648 These failures, which also exist in today's networks, will have a
1649 stronger effect in GNs where a large amount of data will be "stored"
1650 in transit and, to expedite the switching, nodes will apply only
1651 minimal processing to the packets traversing them. In source
1652 routing, for example, a link failure may cause the loss of all
1653 packets sent until the source is notified about the change in
1654 topology. The longer is the delay in recovering from failures, the
1655 higher is the degradation in performance observed by the users.
1656
1657 To minimize the effects of failures, GNs will need to employ error
1658 recovery mechanisms whereby the network detects failures and error
1659 conditions, reconfigures itself to adapt to the new network state,
1660 and notifies peripheral devices of the new configuration. Such
1661 protocols, which have to be developed, will respond quickly, will be
1662 decentralized or distributed to minimize the possibility of fatal
1663 failures, and will complement, rather than replicate, the error
1664 correction mechanisms of the end-to-end protocols, and the two must
1665 operate in coordinated manner. To this end, the peripheral devices
1666 will have to be knowledgeable about the intranet recovery mechanisms
1667 and interact continuously with them to minimize the effect on the
1668 connections they manage.
1669
1670
1671
1672
1673
1674
1675
1676 <span class="grey">Gigabit Working Group [Page 30]</span></pre>
1677 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-31" ></span>
1678 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1679
1680
1681 3.4.5. Flow Control
1682
1683
1684 As networks become faster, two related problems arise. First,
1685 existing flow control mechanisms such as windows do not work well,
1686 because the window must be opened to such an extent to achieve
1687 desired bandwidth that effective flow control cannot be achieved.
1688 Second, especially for long-haul networks, the larger number of bits
1689 in transit at one time becomes so large that most computer messages
1690 will fit into one window. This means that traditional congestion
1691 control schemes will cease to work well.
1692
1693 What is needed is a combination of two approaches, both new. First,
1694 for messages that are small (most messages generated by computers
1695 today will be small, since they will fit into one round-trip time of
1696 future networks), open-loop controls on flow and congestion are
1697 needed. For longer messages (voice or video streams, for example),
1698 some explicit resource commitment will be required.
1699
1700
1701 3.4.6. Latency Control and Real-Time Operations
1702
1703
1704 Currently, there are several distinct approaches to latency control.
1705 First, there are some networks which are physically short, more like
1706 multiprocessor buses. Applications in these networks are built
1707 assuming that delays will be short.
1708
1709 Second, there are networks where the physical length is not
1710 constrained by the design and may differ by orders of magnitude,
1711 depending on the scope of the network. Most general purpose networks
1712 fall in this category. In these networks, one of two things happens.
1713 Either the application takes special steps to deal with variable
1714 latency, such as echo suppression in voice networks, or these
1715 applications are not supported.
1716
1717 For most applications today, the latency in the network is not an
1718 obvious issue so long as the network is not overloaded (which leads
1719 to losses and long queues), because the protocol overhead masks the
1720 variation in the network latency. This balance will change. The
1721 latency due to the speed of light will obviously remain the same, but
1722 the overhead will drop (of necessity if we are to achieve high
1723 performance) which will leave speed of light and queueing as the most
1724 critical sources of delay.
1725
1726 This conclusion implies that if queueing delay can be controlled, it
1727 will be possible to build networks with stable and controlled
1728 latency. If applications exist that require this class of service,
1729
1730
1731
1732 <span class="grey">Gigabit Working Group [Page 31]</span></pre>
1733 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-32" ></span>
1734 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1735
1736
1737 it can be supported. Either the network must be underloaded, so that
1738 queues do not develop at all, or a specific class of service must be
1739 supported in which resources are allocated to stabilize the delay.
1740
1741 If this service is provided, it will still leave the application with
1742 delays that can vary by several orders of magnitude, depending on the
1743 physical size of the network. Research at the application level will
1744 be required to see how applications can be designed to cope with this
1745 variation.
1746
1747
1748 3.4.7. High-Speed Internetworking and Administrational Domains
1749
1750
1751 Internetworking recognized that the value of communication services
1752 increases significantly with wider interconnection but ignored
1753 management and the role of administrations. As a consequence we see
1754 that:
1755
1756 1. The Internet is more or less unmanageable, as evidenced by
1757 performance, reliability, and security problems.
1758
1759 2. The Internet is being stressed by administrators that are
1760 building networks to match their organization rather than the
1761 geography. An example is a set of Ethernets at different
1762 company locations operating as a single Internet network but
1763 geographically dispersed and connected by satellite or leased
1764 lines.
1765
1766 The next generation of internetworking must focus on administration
1767 and management. Internetworking must support cohesion within an
1768 administration and a healthy separation between administrations. To
1769 illustrate by analogy, the American and Soviet embassies in Mexico
1770 City are geographically closer to each other than to their respective
1771 home countries but further in administrational distance, including
1772 security, accounting, etc. The emerging revolution in WANs makes
1773 this issue that much more critical. The amount of communication to
1774 exchange the state of systems is bound to increase enormously. The
1775 potential cost of failures and security violations is frightening.
1776
1777 A promising approach appears to be high-level gateways that guard
1778 between administrations and require negotiations to set up access
1779 paths between administrations. These paths are set up, and labeled
1780 with agreements on authorization, security, accounting, and possible
1781 resource limits. These administrative virtual circuits provide
1782 transparency to the physical and geographical interconnection, but
1783 need not support more than datagram packet delivery. One view is
1784 that of communication contracts with high-level gateways acting as
1785
1786
1787
1788 <span class="grey">Gigabit Working Group [Page 32]</span></pre>
1789 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-33" ></span>
1790 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1791
1792
1793 contract monitors at each end. The key is the focus on controlled
1794 interadministrational connectivity, not the conventional protocol
1795 concerns.
1796
1797 Focus is required on developing an (inter)network management
1798 architecture and the specifics of high-level gateways. The
1799 structures of such gateways will have to take advantage of advances
1800 in multi-processor architectures to handle the processing load.
1801 Moreover, a key issue is being able to optimize communication between
1802 administrations once the contract is in place, but without losing
1803 control. Related is the issue of allowing high-speed interconnection
1804 within a single administration, although geographical dispersed.
1805 Another issue is fault-tolerance. High-level gateways contain state
1806 information whose loss typically disrupts communication. How does
1807 one minimize this problem?
1808
1809 A key goal of these administrational gateways has to be failure
1810 containment: How to protect against external (to administration)
1811 problems and how to prevent local problems imposing liability on
1812 others. A particular area of concern is the self-organizing problems
1813 of large-scale systems, observed by Van Jacobson in the Internet.
1814 Gateways must serve to damp out oscillations and control wide load
1815 swings. Rate control appears to be a key area to investigate as a
1816 basis for buffer management and for congestion control, as well as to
1817 control offered load.
1818
1819 Given the speed of new networks, and the sophistication of the
1820 gateways suggested above, another key area to investigate is the
1821 provision of high-speed network interface adaptors.
1822
1823
1824 3.4.8. Policy-Based Algorithms
1825
1826
1827 Networks of today generally select routes based on minimizing some
1828 measure such as delay. However, in the real world, route selection
1829 will commonly be constrained at the global level by policy issues,
1830 such as access rights to resources and accounting and billing for
1831 usage.
1832
1833 It is difficult for connectionless protocols such as Internet to deal
1834 with policy controls, because a lack of state in the gateway implies
1835 that a separate policy decision must be made for each packet in
1836 isolation. As networks get faster, the cost of this processing will
1837 be intolerable. One possible approach, discussed above, is to move
1838 to a more sophisticated model in which there is knowledge in the
1839 gateways of the ongoing flows. Alternatively, it may be possible to
1840 design gateways that simply cache recent policy evaluations and apply
1841
1842
1843
1844 <span class="grey">Gigabit Working Group [Page 33]</span></pre>
1845 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-34" ></span>
1846 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1847
1848
1849 them to successive packets.
1850
1851 Routing based on policy is particularly difficult because a route
1852 must be globally consistent to be useful; otherwise it may loop.
1853 This implies that the every policy decision must be propagated
1854 globally. Since there can be expected to be a large number of
1855 policies, this global passing of information might easily lead to an
1856 information explosion.
1857
1858 There are at least two solutions. One is to restrict the possible
1859 classes of policy. Another is to use some form of source route, so
1860 that the route consistent with some set of policies is computed at
1861 one point only, and then attached to the packet. Both of these
1862 approaches have problems. A two-pronged research program is needed,
1863 in which mechanisms are proposed, and at the same time the needed
1864 policies are defined.
1865
1866 The same trade-off can be seen for accounting and billing. A single
1867 accounting metric, such as "bytes times distance", could be proposed.
1868 This might be somewhat simple to implement, but would not permit the
1869 definition of individual billing policies, as is now done in the
1870 parts of the telephone system. The current connectionless transport
1871 architectures such as TCP/IP or the connectionless ISO configuration
1872 using TP4 do not have good tools for accounting for traffic, or for
1873 restricting traffic from certain resources. Building these tools is
1874 difficult in a connectionless environment, because an accounting or
1875 control facility must deal with each packet in isolation, which
1876 implies a significant processing burden as part of packet forwarding.
1877 This burden is an increasing problem as switches are expected to
1878 operate faster.
1879
1880 The lack of these tools is proving a significant problem for network
1881 design. Not only are accounting and control needed to support
1882 management requirements, they are needed as a building block to
1883 support enforcement of such things as multiple qualities of service,
1884 as discussed above.
1885
1886 Network accounting is generally considered to be simply a step that
1887 leads to billing, and thus is often evaluated in terms of how simple
1888 or difficult it will be to implement. Yet an accounting and billing
1889 procedure is a mechanism for implementing a policy considered to be
1890 desirable for reasons beyond the scope of accounting per se. For
1891 example, a policy might be established either to encourage or
1892 discourage network use, while fully recovering operational cost. A
1893 policy of encouraging use could be implemented by a relatively high
1894 monthly attachment charge and a relatively low per-packet charge. A
1895 policy of discouraging use could be implemented by a low monthly
1896 charge and a high per-packet charge.
1897
1898
1899
1900 <span class="grey">Gigabit Working Group [Page 34]</span></pre>
1901 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-35" ></span>
1902 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1903
1904
1905 Network administrators have a relatively small number of variables
1906 with which to implement policy objectives. Nevertheless, these
1907 variables can be combined in a number of innovative ways. Some of
1908 the possibilities include:
1909
1910 1. Classes of users (e.g., large or small institutions, for-
1911 profit or non-profit).
1912
1913 2. Classes of service.
1914
1915 3. Time varying (e.g., peak and off-peak).
1916
1917 4. Volume (e.g., volume discounts, or volume surcharges).
1918
1919 5. Access charges (e.g., per port, or port * [bandwidth of
1920 port]).
1921
1922 6. Distance (e.g., circuit-miles, airline miles, number of hops).
1923
1924 Generally, an accounting procedure can be developed to support
1925 voluntary user cooperation with almost any single policy objective.
1926 Difficulties most often arise when there are multiple competing
1927 policy objectives, or when there is no clear policy at all.
1928
1929 Another aspect of accounting and billing procedures which must be
1930 carefully considered is the cost of accumulating and processing the
1931 data on which billing is based. Of particular concern is collection
1932 of detailed data on a per-packet basis. As network circuit data
1933 rates increase, the number of instructions which must be executed on
1934 a per-packet basis can become the limiting factor in system
1935 throughput. Thus, it may be appropriate to prefer accounting and
1936 billing policies and procedures which minimize the difficulty of
1937 collecting data, even if this approach requires a compromise of other
1938 objectives. Similarly, node memory required for data collection and
1939 any network bandwidth required for transmission of the data to
1940 administrative headquarters are factors which must be traded off
1941 against the need to process user packets.
1942
1943
1944 3.4.9. Priority and Preemption
1945
1946
1947 The GN should support multiple levels of priority for traffic and the
1948 preemption of network resources for higher priority use. Network
1949 control traffic should be given the highest priority to ensure that
1950 it is able to pass through the network unimpeded by congestion caused
1951 by user-level traffic. There may be additional military uses for
1952 multiple levels of priority which correspond to rank or level of
1953
1954
1955
1956 <span class="grey">Gigabit Working Group [Page 35]</span></pre>
1957 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-36" ></span>
1958 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
1959
1960
1961 importance of a user or the mission criticality of some particular
1962 data.
1963
1964 The use of and existence of priority levels may be different for
1965 different types of traffic. For example, datagram traffic may not
1966 have multiple priority levels. Because the network's transmission
1967 speed is so high and traffic bursts may be short, it may not make
1968 sense to do any processing in the switches to deal with different
1969 priority levels. Priority will be more important for flow- (or
1970 soft-connection-) oriented data or hard connections in terms of
1971 permitting higher priority connections to be set up ahead of lower
1972 priority connections. Preemption will permit requests for high
1973 priority connections to reclaim network resources currently in use by
1974 lower priority traffic.
1975
1976 Networks such as the Wideband Satellite Network, which supports
1977 datagram and stream traffic, implement four priority levels for
1978 traffic with the highest reserved for network control functions and
1979 the other three for user traffic. The Wideband Network supports
1980 preemption of lower priority stream allocations by higher priority
1981 requests. An important component of the use of priority and
1982 preemption is the ability to notify users when requests for service
1983 have been denied, or allocations have been modified or disrupted.
1984 Such mechanisms have been implemented in the Wideband Network for
1985 streams and dynamic multicast groups.
1986
1987 Priority and preemption mechanisms for a GN will have to be
1988 implemented in an extremely simple way so that they can take effect
1989 very quickly. It is likely that they will have to built into the
1990 hardware of the switch fabric.
1991
1992
1993 3.5. User and Network Services
1994
1995
1996 As discussed in <a href="#section-2">Section 2</a> above, there will need to be certain
1997 services provided as part of the network operation to the users
1998 (people) themselves and to the machines that connect to the network.
1999 These services, which include such capabilities as white and yellow
2000 pages (allowing users to determine what the appropriate network
2001 identification is for other users and for network-available computing
2002 resources) and distributed fault identification and isolation, are
2003 needed in current networks and will continue to be required in the
2004 networks of the future. The speed of the GN will serve to accentuate
2005 this requirement, but at the same time will allow for new
2006 architectures to be put in place for such services. For example,
2007 Ethernet speeds in the local environment have allowed for more usable
2008 services to be provided.
2009
2010
2011
2012 <span class="grey">Gigabit Working Group [Page 36]</span></pre>
2013 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-37" ></span>
2014 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
2015
2016
2017 3.5.1. Impact of High Bandwidth
2018
2019
2020 One issue that will need to be addressed is the impact on the user of
2021 such high-bandwidth capabilities. Users are already becoming
2022 saturated by information in the modern information-rich environment.
2023 (Many of us receive more than 50 electronic mail messages each day,
2024 each requiring some degree of human attention.) Methods will be
2025 needed to allow users to cope with this ever-expanding access to
2026 data, or we will run the risk of users turning back to the relative
2027 peace and quiet of the isolated office.
2028
2029
2030 3.5.2. Distributed Network Directory
2031
2032
2033 A distributed network directory can support the user-level directory
2034 services and the lower-level name-to-address mapping services
2035 described elsewhere in this report. It can also support distributed
2036 systems and network management facilities by storing additional
2037 information about named objects. For example, the network directory
2038 might store node configurations or security levels.
2039
2040 Distributing the directory eases and decentralizes the administrative
2041 burdens and provides a more robust and survivable implementation.
2042
2043 One approach toward implementing a distributed network directory
2044 would be to base it upon the CCITT X.500/ISO DIS 9594 standard. This
2045 avoids starting from ground zero and has the advantage of
2046 facilitating interoperability with other communications networks.
2047 However, research and development will be required even if this path
2048 is chosen.
2049
2050 One area in which research and development are required is in the
2051 services supplied by the distributed network directory. The X.500
2052 standard is very general and powerful, but so far specific provisions
2053 have been made only for storing information about network users and
2054 applications. As mentioned elsewhere, multilevel security is not
2055 addressed by X.500, and the approach taken toward authentication must
2056 be carefully considered in view of DoD requirements. Also, X.500
2057 assumes that administration of the directory will be done locally and
2058 without the need for standardization; this may not be true of GN or
2059 the larger national research network.
2060
2061 The model and algorithms used by a distributed network directory
2062 constitute a second area of research. The model specified by X.500
2063 must be extended into a framework that provides the necessary
2064 flexibility in terms of services, responsiveness, data management
2065
2066
2067
2068 <span class="grey">Gigabit Working Group [Page 37]</span></pre>
2069 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-38" ></span>
2070 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
2071
2072
2073 policies, and protocol layer utilization. Furthermore, the internal
2074 algorithms and mechanisms of X.500 must be extended in a number of
2075 areas; for example, to support redundancy of the X.500 database,
2076 internal consistency checking, fuller sharing of information about
2077 the distribution of data, and defined access-control mechanisms.
2078
2079
2080 4. Avenues of Approach
2081
2082
2083 Ongoing research and commercial activities provide an opportunity for
2084 more rapidly attacking some of the above research issues. At the
2085 same time, there needs to be attention paid to the overall technical
2086 approach used to allow multiple potential solutions to be explored
2087 and allow issues to be attacked in parallel.
2088
2089
2090 4.1. Small Prototype vs. Nationwide Network
2091
2092
2093 The central question is how far to jump, and how far can the current
2094 approaches get. That is, how far will connectionless network service
2095 get us, how far will packet switching get us, and how far do we want
2096 to go. If our goal is a Gbit/s net, then that is what we should
2097 build. Building a 100 Mbit/s network to achieve a GN is analogous to
2098 climbing a tree to get to the moon. It may get you closer, but it
2099 will never get you there.
2100
2101 There are currently some network designs which can serve as the basis
2102 for a GN prototype. The next step is some work by experts in
2103 photonics and possibly high-speed electronics to explore ease of
2104 implementation. Developing a prototype 3-5 node network at a Gbit/s
2105 data rate is realistic at this point and would demonstrate wide-area
2106 (40 km or more) Gbit/s networking.
2107
2108 DARPA should consider installing a Gbit/s cross-country set of
2109 connected links analogous to the NSF backbone in 2 years. A Gbit/s
2110 link between the east and west coasts would open up a whole new
2111 generation of (C3I), distributed computing, and parallel computing
2112 research possibilities and would reestablish DARPA as the premier
2113 network research funding agency in the country. This will require
2114 getting "dark" fiber from one or more of the common carriers and some
2115 collaboration with these organizations on repeaters, etc. With this
2116 collaboration, the time to a commercial network in the Gbit/s range
2117 would be substantially reduced, and the resulting nationwide GN would
2118 give the United States an enormous technical and economic advantage
2119 over countries without it.
2120
2121
2122
2123
2124 <span class="grey">Gigabit Working Group [Page 38]</span></pre>
2125 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-39" ></span>
2126 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
2127
2128
2129 Demonstrating a high-bandwidth WAN is not enough, however. As one
2130 can see from the many research issues identified above, it will be
2131 necessary to pursue via study and experiment the issues involved in
2132 interconnecting high-bandwidth networks into a high-bandwidth
2133 internet. These experiments can be done through use of a new
2134 generation of internet, even if it requires starting at lower speeds
2135 (e.g., T1 through 100 Mbit/s). Appropriate care must be given,
2136 however, to assure that the capabilities that are demonstrated are
2137 applicable to the higher bandwidths (Gbit/s) as they emerge.
2138
2139
2140 4.2. Need for Parallel Efforts/Approaches
2141
2142
2143 Parallel efforts will therefore be required for two major reasons.
2144 First is the need to pursue alternative approaches (e.g., different
2145 strategies for high-bandwidth switching, different addressing
2146 techniques, etc). This is the case for most research programs, but
2147 it is made more difficult here by the costs of prototyping. Thus, it
2148 is necessary that appropriate review take place in the decisions as
2149 to which efforts are supported through prototyping.
2150
2151 In addition, it will be necessary to pursue the different aspects of
2152 the program in parallel. It will not be possible to wait until the
2153 high-bandwidth network is available before starting on prototyping
2154 the high-bandwidth internet. Thus, a phased and evolutionary
2155 approach will be needed.
2156
2157
2158 4.3. Collaboration with Common Carriers
2159
2160
2161 Computer communication networks in the United States today
2162 practically ignore the STN (the Switched Telephone Network), except
2163 for buying raw bandwidth through it. However, advances in network
2164 performance are based on improvements in the underlying communication
2165 media, including satellite communication, fiber optics, and photonic
2166 switching.
2167
2168 In the past we used "their" transmission under "our" switching. An
2169 alternative approach is to utilize the common-carrier switching
2170 capabilities as an integral part of the networking architecture. We
2171 must take an objective scientific and economic look and reevaluate
2172 this question.
2173
2174 Another place for cooperation with the common carriers is in the area
2175 of network addressing. Their addressing scheme ("numbering plan")
2176 has a few advantages such as proven service to 300 million users [<a href="#ref-4" title="000 users (less than 1:30">4</a>].
2177
2178
2179
2180 <span class="grey">Gigabit Working Group [Page 39]</span></pre>
2181 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-40" ></span>
2182 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
2183
2184
2185 On the other hand, the common carriers have far fewer administrative
2186 domains (area codes) than the current plethora of locally
2187 administered local area networks in the internet system.
2188
2189 It is likely that future networks will eventually be managed and
2190 operated by commercial communications providers. A way to maximize
2191 technology transfer from the research discussed here to the
2192 marketplace is to involve the potential carriers from the start.
2193 However, it is not clear that the goals of commercial communications
2194 providers, who have typically been most interested in meeting the
2195 needs of 90+ percent of the user base, will be compatible with the
2196 goals of the research described here. Thus, while we recommend that
2197 the research program involve an appropriate amalgam of academia and
2198 industry, paying particular attention to involvement of the potential
2199 system developers and operators, we also caution that the specific
2200 and unique goals of the DARPA program must be retained.
2201
2202
2203 4.4. Technology Transfer
2204
2205
2206 As we said above, it is our belief that future networks will
2207 ultimately be managed and operated by commercial communications
2208 providers. (Note that this may not be the common carriers as we know
2209 them today, but may be value-added networks using common carrier
2210 facilities.) The way to assure technology transfer, in our belief, is
2211 to involve the potential system developers from the start. We
2212 therefore believe that the research program would benefit from an
2213 appropriate amalgam of university and industry, with provision for
2214 close involvement of the potential system developers and operators.
2215
2216
2217 4.5. Standards
2218
2219
2220 The Internet program was a tremendous success in influencing national
2221 and international standards. While there were changes to the
2222 protocols, the underlying technology and approaches used by CCITT and
2223 ISO in the standardization of packet-switched networks clearly had
2224 its roots in the DARPA internet. Nevertheless, this has had some
2225 negative impact on the research program, as the evolution of the
2226 standards led to pressure to adopt them in the research environment.
2227
2228 Thus, it appears that there is a "catch-22" here. It is desirable
2229 for the technology base developed in the research program to have
2230 maximal impact on the standards activities. This is expedited by
2231 doing the research in the context of the standards environment.
2232 However, standards by their very nature will always lag behind the
2233
2234
2235
2236 <span class="grey">Gigabit Working Group [Page 40]</span></pre>
2237 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-41" ></span>
2238 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
2239
2240
2241 research environment.
2242
2243 The only reasonable approach, therefore, appears to be an occasional
2244 "checkpointing" of the research environment, where the required
2245 conversions take place to allow a new plateau of standards to be used
2246 for future evolution and research. A good example is conducting
2247 future research in mail using X.400 and X.500 where possible.
2248
2249
2250 5. Conclusions
2251
2252
2253 We hope that this document has provided a useful compendium of those
2254 research issues critical to achieving the FCCSET phase III
2255 recommendations. These problems interact in a complex way. If the
2256 only goal of a new network architecture was high speed, reasonable
2257 solutions would not be difficult to propose. But if one must achieve
2258 higher speeds while supporting multiple services, and at the same
2259 time support the establishment of these services across
2260 administrative boundaries, so that policy concerns (e.g., access
2261 control) must be enforced, the interactions become complex.
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292 <span class="grey">Gigabit Working Group [Page 41]</span></pre>
2293 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-42" ></span>
2294 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
2295
2296
2297 APPENDIX
2298
2299 <span class="h2"><a class="selflink" id="appendix-A" href="#appendix-A">A</a>. Current R and D Activities</span>
2300
2301 In this appendix, we provide pointers to some ongoing activities in
2302 the research and development community of which the group was aware
2303 relevant to the goal of achieving the GN. In some cases, a short
2304 abstract is provided of the research. Neither the order of the
2305 listing (which is random) nor the amount of detail provided is meant
2306 to indicate in any way the significance of the activity. We hope
2307 that this set of pointers will be useful to anyone who chooses to
2308 pursue the research issues discussed in this report.
2309
2310 1. Grumman (at Bethpage) is working on a three-year DARPA
2311 contract, started in January 1988 to develop a 1.6 Gbit/s LAN,
2312 for use on a plane or ship, or as a "building block". It is
2313 really raw transport capacity running on two fibers in a
2314 token-ring like mode. First milestone (after one year?) is to
2315 be a 100 Mbit/s demonstration.
2316
2317 2. BBN Laboratories, as part of its current three-year DARPA
2318 Network-Oriented Systems contract, has proposed design
2319 concepts for a 10-100 Gbit/s wide area network. Work under
2320 this effort will include wavelength division multiplexing,
2321 photonic switching, self-routing packets, and protocol design.
2322
2323 3. Cheriton (Stanford) research on Blazenet, a high-bandwidth
2324 network using photonic switching.
2325
2326 4. Acampora (Bell Labs) research on the use of wavelength
2327 division multiplexing for building a shared optical network.
2328
2329 5. Yeh is reserching a VLSI approach to building high-bandwidth
2330 parallel processing packet switch.
2331
2332 6. Bell Labs is working on a Metropolitan Area Network called
2333 "Manhattan Street Net." This work, under Dr. Maxemchuck, is
2334 similar to Blazenet. It is in the prototype stage for a small
2335 number of street intersections; ultimately it is meant to be
2336 city-wide. Like Blazenet, is uses photonic switching 2 x 2
2337 lithium niobate block switches.
2338
2339 7. Ultra Network Technologies is a Silicon Valley company which
2340 has a (prototype) Gbit/s fiber link which connects backplanes.
2341 This is based on the ISO-TP4 transport protocol.
2342
2343 8. Jonathan Turner, Washington University, is working on a
2344 Batcher-Banyan Multicast Net, based on the "SONET" concept,
2345
2346
2347
2348 <span class="grey">Gigabit Working Group [Page 42]</span></pre>
2349 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-43" ></span>
2350 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
2351
2352
2353 which provides 150 Mbit/s per pipe.
2354
2355 9. David Sincowskie, Bellcore, is working with Batcher-Banyan
2356 design and has working 32x32 switches.
2357
2358 10. Stratacom has a commercial product which is really a T1 voice
2359 switch implemented internally by a packet switch, where the
2360 packet is 192 bits (T1 frame). This switch can pass 10,000
2361 packets per second.
2362
2363 11. Stanford NAB provides 30-50 Mbit/s throughput on 100 Mbit/s
2364 connection using Versatile Message Transaction Protocol (VMTP)
2365 [see <a href="./rfc1045">RFC 1045</a>]
2366
2367 12. The December issue of IEEE Journal on Selected Areas in
2368 Communications, provides much detail concerning interconnects.
2369
2370 13. Ultranet Technology has a 480 Mbit/s connection using modified
2371 ISO TP4.
2372
2373 14. At MIT, Dave Clark has an architecture proposal of interest.
2374
2375 15. At CMU, the work of Eric Cooper is relevant.
2376
2377 16. At Protocol Engines, Inc., Greg Chesson is working on an XTP-
2378 based system.
2379
2380 17. Larry Landweber at Wisconsin University is doing relevant
2381 work.
2382
2383 18. Honeywell is doing relevant work for NASA.
2384
2385 19. Kung at CMU is working on a system called "Nectar" based on a
2386 STARLAN on fiber connecting dissimilar processors.
2387
2388 20. Burroughs (now Unisys) has some relevant work within the IEEE
2389 802.6 committee.
2390
2391 21. Bellcore work in "Switched Multimedia Datanet Service" (SMDS)
2392 is relevant (see paper supplied by Dave Clark).
2393
2394 22. FDDI-2, a scheme for making TDMA channel allocations at 200
2395 Mbit/s.
2396
2397 23. NRI, Kahn-Farber Proposal to NSF, is a paper design for high-
2398 bandwidth network.
2399
2400 24. Barry Goldstein work, IBM-Yorktown.
2401
2402
2403
2404 <span class="grey">Gigabit Working Group [Page 43]</span></pre>
2405 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-44" ></span>
2406 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
2407
2408
2409 25. Bell Labs S-Net, 1280 Mbit/s prototype.
2410
2411 26. Fiber-LAN owned by Bell South and SECOR, a pre-prototype 575
2412 Mbit/s Metro Area Net.
2413
2414 27. Bellcore chip implementation of FASTNET (1.2 Gbit/s).
2415
2416 28. Scientific Computer Systems, San Diego, 1.4 Gbit/s prototype.
2417
2418 29. BBN Monarch Switch, Space Division pre-prototype, chips being
2419 fabricated, 64 Mbit/s per path.
2420
2421 30. Proteon, 80 Mbit/s token ring.
2422
2423 31. Toronto University, 150 Mbit/s "tree"--- really a LAN.
2424
2425 32. NSC Hyperchannel II, reputedly available at 250 Mbit/s.
2426
2427 33. Tobagi at Stanford working on EXPRESSNET; not commercially
2428 available.
2429
2430 34. Columbia MAGNET-- 150 Mbit/s.
2431
2432 35. Versatile Message Transaction Protocol (VMTP).
2433
2434 36. ST integrated with IP.
2435
2436 37. XTP (Chesson).
2437
2438 38. Stanford Transport Gateway.
2439
2440 39. X.25/X.75.
2441
2442 40. Work of the Internet Activities Board.
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460 <span class="grey">Gigabit Working Group [Page 44]</span></pre>
2461 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-45" ></span>
2462 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
2463
2464
2465 <span class="h2"><a class="selflink" id="appendix-B" href="#appendix-B">B</a>. Gigabit Working Group Members</span>
2466
2467 Member Affiliation
2468
2469 Gordon Bell Ardent Computers
2470 Steve Blumenthal BBN Laboratories
2471 Vint Cerf Corporation for National Research Initiatives
2472 David Cheriton Stanford University
2473 David Clark Massachusetts Institute of Technology
2474 Barry Leiner (Chairman) Research Institute for Advanced Computer Science
2475 Robert Lyons Defense Communication Agency
2476 Richard Metzger Rome Air Development Center
2477 David Mills University of Delaware
2478 Kevin Mills National Bureau of Standards
2479 Chris Perry MITRE
2480 Jon Postel USC Information Sciences Institute
2481 Nachum Shacham SRI International
2482 Fouad Tobagi Stanford University
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516 <span class="grey">Gigabit Working Group [Page 45]</span></pre>
2517 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-46" ></span>
2518 <span class="grey"><a href="./rfc1077">RFC 1077</a> November 1988</span>
2519
2520
2521 End Notes
2522
2523 [<a id="ref-1">1</a>] Workshop on Computer Networks, 17-19 February 1987, San Diego,
2524 CA.
2525
2526 [<a id="ref-2">2</a>] "A Report to the Congress on Computer Networks to Support
2527 Research in the United States: A Study of Critical Problems and
2528 Future Options", White House Office of Scientific and Technical
2529 Policy (OSTP), November 1987.
2530
2531 [<a id="ref-3">3</a>] We distinguish in the report between development of a backbone
2532 network providing gigabit capacity, the GB, and an
2533 interconnected set of high-speed networks providing high-
2534 bandwidth service to the user, the Gigabit Network (GN).
2535
2536 [<a id="ref-4">4</a>] Incidentally, they already manage to serve 150 million
2537 subscribers in an 11-digit address-space (about 1:600 ratio).
2538 We have a 9.6-digit address-space and are running into troubles
2539 with much less than 100,000 users (less than 1:30,000 ratio).
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572 Gigabit Working Group [Page 46]
2573 </pre>