Debian Codegraph @

rfc1125.html

raw
1 <pre>Network Working Group D. Estrin
2 Request for Comments: 1125 USC Computer Science Department
3 November 1989
4
5
6 <span class="h1">POLICY REQUIREMENTS FOR INTER ADMINISTRATIVE DOMAIN ROUTING</span>
7
8 <span class="h2"><a class="selflink" id="section-1" href="#section-1">1</a> STATUS OF THIS MEMO</span>
9
10 The purpose of this memo is to focus discussion on particular
11 problems in the Internet and possible methods of solution. No
12 proposed solutions in this document are intended as standards for the
13 Internet. Rather, it is hoped that a general consensus will emerge
14 as to the appropriate solution to such problems, leading eventually
15 to the development and adoption of standards. Distribution of this
16 memo is unlimited.
17
18 <span class="h2"><a class="selflink" id="section-2" href="#section-2">2</a> ABSTRACT</span>
19
20 Efforts are now underway to develop a new generation of routing
21 protocol that will allow each Administrative Domain (AD) in the
22 growing Internet (and internets in general) to independently express
23 and enforce policies regarding the flow of packets to, from, and
24 through its resources. (FOOTNOTE 1: The material presented here
25 incorporates discussions held with members of the IAB Autonomous
26 Networks Research Group and the Open Routing Working Group.) This
27 document articulates the requirements for policy based routing and
28 should be used as input to the functional specification and
29 evaluation of proposed protocols.
30
31 Two critical assumptions will shape the type of routing mechanism
32 that is devised: (1) the topological organization of ADs, and (2) the
33 type and variability of policies expressed by ADs. After justifying
34 our assumptions regarding AD topology we present a taxonomy, and
35 specific examples, of policies that must be supported by a PR
36 protocol. We conclude with a brief discussion of policy routing
37 mechanisms proposed in previous RFCs (827, 1102, 1104, 1105). Future
38 RFCs will elaborate on the architecture and protocols needed to
39 support the requirements presented here.
40
41 <span class="h2"><a class="selflink" id="section-3" href="#section-3">3</a> BACKGROUND</span>
42
43 The Research Internet has evolved from a single backbone wide area
44 network with many connected campus networks, to an internet with
45 multiple cross-country backbones, regional access networks, and a
46 profusion of campus networks. (FOOTNOTE 2: The term Research Internet
47 refers to a collection of government, university, and some private
48 company, networks that are used by researchers to access shared
49
50
51
52 <span class="grey">Estrin [Page 1]</span></pre>
53 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-2" ></span>
54 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
55
56
57 computing resources (e.g., supercomputers), and for research related
58 information exchange (e.g., distribution of software, technical
59 documents, and email). The networks that make up the Research
60 Internet run the DOD Internet Protocol [<a href="#ref-1" title="Network Information Center">1</a>].) At times during its
61 development the Research Internet topology appeared somewhat chaotic.
62 Overlapping facilities and lateral (as opposed to hierarchical)
63 connections seemed to be the rule rather than the exception. Today
64 the Research Internet topology is becoming more regular through
65 coordination of agency investment and adoption of a hierarchy similar
66 to that of the telephone networks'. The result is several
67 overlapping wide area backbones connected to regional networks, which
68 in turn connect to campus networks at universities, research
69 laboratories, and private companies. However, the telephone network
70 has lateral connections only at the highest level, i.e., between long
71 haul carriers. In the Research Internet there exist lateral
72 connections at each level of the hierarchy, i.e., between campus (and
73 regional) networks as well.
74
75 Additional complexity is introduced in the Research Internet by
76 virtue of connections to private networks. Many private companies are
77 connected to the Research Internet for purposes of research or
78 support activities. These private companies connect in the same
79 manner as campuses, via a regional network or via lateral links to
80 other campuses. However, many companies have their own private wide
81 area networks which physically overlap with backbone and/or regional
82 networks in the research internet, i.e., private vertical bypass
83 links.
84
85 Implicit in this complex topology are organizational boundaries.
86 These boundaries define Administrative Domains (ADs) which preclude
87 the imposition of a single, centralized set of policies on all
88 resources. The subject of this paper is the policy requirements for
89 resource usage control in the Research Internet.
90
91 In the remainder of this section we describe the policy routing
92 problem in very general terms. <a href="#section-4">Section 4</a> examines the constraints and
93 requirements that makes the problem challenging, and leads us to
94 conclude that a new generation of routing and resource control
95 protocols are needed. <a href="#section-5">Section 5</a> provides more detail on our
96 assumptions as to the future topology and configuration of
97 interconnected ADs. We return to the subject of policy requirements
98 in <a href="#section-7">Section 7</a> and categorize the different types of policies that ADs
99 in the research internet may want to enforce. Included in this
100 section are examples of FRICC policy statements. (FOOTNOTE 3: The
101 Federal Research Internet Coordinating Committee (FRICC) is made up
102 of representatives of each of the major agencies that are involved in
103 networking. They have been very effective in coordinating their
104 efforts to eliminate inefficient redundancy and have proposed a plan
105
106
107
108 <span class="grey">Estrin [Page 2]</span></pre>
109 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-3" ></span>
110 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
111
112
113 for the next 10 years of internetworking for the government,
114 scientific, and education community [<a href="#ref-2" title="The Federal Research Internet Coordinating Committee and National Research Network">2</a>].) <a href="#section-7">Section 7</a> identifies types
115 of policy statements that are problematic to enforce due to their
116 dynamics, granularity, or performance implications. Several proposed
117 mechanisms for supporting PR (including RFCs 827, 1102, 1104, 1105)
118 are discussed briefly in <a href="#section-8">Section 8</a>. Future RFCs will elaborate on the
119 architecture and protocols needed to support the requirements
120 presented here.
121
122 <span class="h3"><a class="selflink" id="section-3.1" href="#section-3.1">3.1</a> POLICY ROUTING</span>
123
124 Previous protocols such as the Exterior Gateway Protocol (EGP)[<a href="#ref-3" title="Exterior Gateway Protocol (EGP)">3</a>]
125 embodied a limited notion of policy and ADs. In particular,
126 autonomous system boundaries constrained the flow of routing database
127 information, and only indirectly affected the flow of packets
128 themselves. We consider an Administrative Domain (AD) to be a set of
129 hosts and network resources (gateways, links, etc.) that is governed
130 by common policies. In large internets that cross organization
131 boundaries, e.g., the Research Internet, inter-AD routes must be
132 selected according to policy-related parameters such as cost and
133 access rights, in addition to the traditional parameters of
134 connectivity and congestion. In other words, Policy Routing (PR) is
135 needed to navigate through the complex web of policy boundaries
136 created by numerous interconnected ADs. Moreover, each AD has its own
137 privileges and perspective and therefore must make its own evaluation
138 of legal and preferred routes. Efforts are now underway to develop a
139 new generation of routing protocol that will allow each AD to
140 independently express and enforce policies regarding the flow of
141 packets to, from, and through its resources [<a href="#ref-4" title="Policy Routing in Internet Protocols">4</a>]. (FOOTNOTE 4: These
142 issues are under investigation by the IAB Autonomous Networks
143 Research Group and the IAB Open Routing Working Group. For further
144 information contact the author.)
145
146 The purpose of this paper is to articulate the requirements for such
147 policy based routing. Two critical assumptions will shape the type of
148 routing mechanism that is devised:
149
150 * The topological organization of ADs, and
151 * The type and variability of policies expressed by ADs.
152
153 We make use of the policies expressed by owners of current Research
154 Internet resources and private networks connected to the Research
155 Internet to generalize types of policies that must be supported. This
156 top down effort must be done with attention to the technical
157 implications of the policy statements if the result is to be useful
158 in guiding technical development. For example, some ADs express the
159 desire to enforce local constraints over how packets travel to their
160 destination. Other ADs are only concerned with preventing use of
161
162
163
164 <span class="grey">Estrin [Page 3]</span></pre>
165 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-4" ></span>
166 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
167
168
169 their own network resources by restricting transit. Still other ADs
170 are concerned primarily with recovering the expense of carrying
171 traffic and providing feedback to users so that users will limit
172 their own data flows; in other words they are concerned with
173 charging. We refer to ADs whose primary concern is communication to
174 and from hosts within their AD as stub and to ADs whose primary
175 concern is carrying packets to and from other ADs as transit}. If we
176 address control of transit alone, for example, the resulting
177 mechanisms will not necessarily allow an AD to control the flow of
178 its packets from source to destination, or to implement flexible
179 charging schemes. (FOOTNOTE 5: Gene Tsudik uses the analogy of
180 international travel to express the need for source and transit
181 controls. Each country expresses its own policies about travel to and
182 through its land. Travel through one country enroute to another is
183 analogous to transit traffic in the network world. A traveler
184 collects policy information from each of the countries of interest
185 and plans an itinerary that conforms to those policies as well as the
186 preferences of the traveler and his/her home nation. Thus there is
187 both source and transit region control of routing.) Our purpose is
188 to articulate a comprehensive set of requirements for PR as input to
189 the functional specification, and evaluation, of proposed protocols.
190
191 <span class="h2"><a class="selflink" id="section-4" href="#section-4">4</a> WHY THE PROBLEM IS DIFFICULT</span>
192
193 Before proceeding with our description of topology and policy
194 requirements this section outlines several assumptions and
195 constraints, namely: the lack of global authority, the need to
196 support network resource sharing as well as network interconnection,
197 the complex and dynamic mapping of users to ADs and privileges, and
198 the need for accountability across ADs. These assumptions limit the
199 solution space and raise challenging technical issues.
200
201 The purpose of policy based routing is to allow ADs to interconnect
202 and share computer and network resources in a controlled manner.
203 Unlike many other problems of resource control, there is no global
204 authority. Each AD defines its own policies with respect to its own
205 traffic and resources. However, while we assume no global authority,
206 and no global policies, we recognize that complete autonomy implies
207 no dependence and therefore no communication. The multi-organization
208 internets addressed here have inherent regions of autonomy, as well
209 as requirements for interdependence. Our mechanisms should allow ADs
210 to design their boundaries, instead of requiring that the boundaries
211 be either impenetrable or eliminated.
212
213 One of the most problematic aspects of the policy routing
214 requirements identified here is the need to support both network
215 resource sharing and interconnection across ADs. An example of
216 resource sharing is two ADs (e.g., agencies, divisions, companies)
217
218
219
220 <span class="grey">Estrin [Page 4]</span></pre>
221 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-5" ></span>
222 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
223
224
225 sharing network resources (e.g., links, or gateways and links) to
226 take advantage of economies of scale. Providing transit services to
227 external ADs is another example of network resource sharing.
228 Interconnection is the more common example of ADs interconnecting
229 their independently used network resources to achieve connectivity
230 across the ADs, i.e., to allow a user in one AD to communicate with
231 users in another AD. In some respects, network resource control is
232 simpler than network interconnection control since the potential
233 dangers are fewer (i.e., denial of service and loss of revenue as
234 compared with a wide range of attacks on end systems through network
235 interconnection). However, controlled network resource sharing is
236 more difficult to support. In an internet a packet may travel
237 through a number of transit ADs on its way to the destination.
238 Consequently, policies from all transit ADs must be considered when a
239 packet is being sent, whereas for stub-AD control only the policies
240 of the two end point ADs have to be considered. In other words,
241 controlled network resource sharing and transit require that policy
242 enforcement be integrated into the routing protocols themselves and
243 can not be left to network control mechanisms at the end points.
244 (FOOTNOTE 6&amp;7: Another difference is that in the interconnect case,
245 traffic traveling over AD A's network resources always has a member
246 of AD A as its source or destination (or both). Under resource
247 sharing arrangements members of both AD A and B are connected to the
248 same resources and consequently intra-AD traffic (i.e., packets
249 sourced and destined for members of the same AD) travels over the
250 resources. This distinction is relevant to the writing of policies in
251 terms of principal affiliation. Economies of scale is one motivation
252 for resource sharing. For example, instead of interconnecting
253 separately to several independent agency networks, a campus network
254 may interconnect to a shared backbone facility. Today,
255 interconnection is achieved through a combination of AD specific and
256 shared arrangements. We expect this mixed situation to persist for
257 "well-connected" campuses for reasons of politics, economics, and
258 functionality (e.g., different characteristics of the different
259 agency-networks). See <a href="#section-5">Section 5</a> for more discussion.)
260
261 Complications also result from the fact that legitimate users of an
262 AD's resources are not all located in that AD. Many users (and their
263 computers) who are funded by, or are affiliated with, a particular
264 agency's program reside within the AD of the user's university or
265 research laboratory. They reside in a campus AD along with users who
266 are legitimate users of other AD resources. Moreover, any one person
267 may be a legitimate user of multiple AR resources under varying
268 conditions and constraints (see examples in <a href="#section-6">Section 6</a>). In addition,
269 users can move from one AD to another. In other words, a user's
270 rights can not be determined solely based on the AD from which the
271 user's communications originate. Consequently, PR must not only
272 identify resources, it must identify principals and associate
273
274
275
276 <span class="grey">Estrin [Page 5]</span></pre>
277 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-6" ></span>
278 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
279
280
281 different capabilities and rights with different principals. (The
282 term principal is taken from the computer security community[7].)
283
284 One way of reducing the compromise of autonomy associated with
285 interconnection is to implement mechanisms that assure
286 accountability} for resources used. Accountability may be enforced a
287 priori, e.g., access control mechanisms applied before resource usage
288 is permitted. Alternatively, accountability may be enforced after
289 the fact, e.g., record keeping or metering that supports detection
290 and provides evidence to third parties (i.e., non-repudiation).
291 Accountability mechanisms can also be used to provide feedback to
292 users as to consumption of resources. Internally an AD often decides
293 to do away with such feedback under the premise that communication is
294 a global good and should not be inhibited. There is not necessarily a
295 "global good" across AD boundaries. Therefore, it becomes more
296 appropriate to have resource usage visible to users, whether or not
297 actual charging for usage takes place. Another motivation that
298 drives the need for accountability across AD boundaries is the
299 greater variability in implementations. Different implementations of
300 a single network protocol can vary greatly as to their efficiency
301 [<a href="#ref-8" title="Palo Alto">8</a>]. We can not assume control over implementation across AD
302 boundaries. Feedback mechanisms such as metering (and charging in
303 some cases) would introduce a concrete incentive for ADs to employ
304 efficient and correct implementations. PR should allow an AD to
305 advertise and apply such accounting measures to inter-AD traffic.
306
307 In summary, the lack of global authority, the need to support network
308 resource sharing as well as network interconnection, the complex and
309 dynamic mapping of users to ADs and rights, and the need for
310 accountability across ADs, are characteristics of inter-AD
311 communications which must be taken into account in the design of both
312 policies and supporting technical mechanisms.
313
314 <span class="h2"><a class="selflink" id="section-5" href="#section-5">5</a> TOPOLOGY MODEL OF INTERNET</span>
315
316 Before discussing policies per se, we outline our model of inter-AD
317 topology and how it influences the type of policy support required.
318 Most members of the Internet community agree that the future Internet
319 will connect on the order of 150,000,000 termination points and
320 100,000 ADs. However, there are conflicting opinions as to the AD
321 topology for which we must design PR mechanisms. The informal
322 argument is described here.
323
324 SIMPLE AD TOPOLOGY AND POLICY MODEL Some members of the Internet
325 community believe that the current complex topology of interconnected
326 ADs is a transient artifact resulting from the evolutionary nature of
327 the Research Internet's history. (FOOTNOTE 9: David Cheriton of
328 Stanford University articulated this side of the argument at an
329
330
331
332 <span class="grey">Estrin [Page 6]</span></pre>
333 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-7" ></span>
334 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
335
336
337 Internet workshop in Santa Clara, January, 1989). The critical points
338 of this argument relate to topology and policy. They contend that in
339 the long term the following three conditions will prevail:
340
341 * The public carriers will provide pervasive, competitively
342 priced, high speed data services.
343
344 * The resulting topology of ADs will be
345 stub (not transit) ADs connected to regional
346 backbones, which in turn interconnect via multiple,
347 overlapping long haul backbones, i.e., a hierarchy with
348 no lateral connections between stub-ADs or regionals,
349 and no vertical bypass links.
350
351 * The policy requirements of the backbone and stub-ADs
352 will be based only on charging for resource usage at the
353 stub-AD to backbone-AD boundary, and to settling accounts
354 between neighboring backbone providers (regional to long haul,
355 and long haul to long haul).
356
357 Under these assumptions, the primary requirement for general AD
358 interconnect is a metering and charging protocol. The routing
359 decision can be modeled as a simple least cost path with the metric
360 in dollars and cents. In other words, restrictions on access to
361 transit services will be minimal and the functionality provided by
362 the routing protocol need not be changed significantly from current
363 day approaches.
364
365 COMPLEX AD TOPOLOGY AND POLICY MODEL The counter argument is that a
366 more complex AD topology will persist. (FOOTNOTE 10: Much of the
367 remainder of this paper attempts to justify and provide evidence for
368 this statement.) The different assumptions about AD topology lead to
369 the significantly different assumptions about AD policies.
370
371 This model assumes that the topology of ADs will in many respects
372 agree with the previous model of increased commercial carrier
373 participation and resulting hierarchical structure. However, we
374 anticipate unavoidable and persistent exceptions to the hierarchy.
375 We assume that there will be a relatively small number of long haul
376 transit ADs (on the order of 100), but that there may be tens of
377 thousands of regional ADs and hundreds of thousands of stub ADs
378 (e.g., campuses, laboratories, and private companies). The competing
379 long haul offerings will differ, both in the services provided and in
380 their packaging and pricing. Regional networks will overlap less and
381 will connect campus and private company networks. However, many
382 stub-ADs will retain some private lateral links for political,
383 technical, and reliability reasons. For example, political
384 incentives cause organizations to invest in bypass links that are not
385
386
387
388 <span class="grey">Estrin [Page 7]</span></pre>
389 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-8" ></span>
390 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
391
392
393 always justifiable on a strict cost comparison basis; specialized
394 technical requirements cause organizations to invest in links that
395 have characteristics (e.g., data rate, delay, error, security) not
396 available from public carriers at a competitive rate; and critical
397 requirements cause organizations to invest in redundant back up links
398 for reliability reasons. These exceptions to the otherwise regular
399 topology are not dispensible. They will persist and must be
400 accommodated, perhaps at the expense of optimality; see <a href="#section-5">Section 5</a> for
401 more detail. In addition, many private companies will retain their
402 own private long haul network facilities. (FOOTNOTE 11: While
403 private voice networks also exist, private data networks are more
404 common. Voice requirements are more standardized because voice
405 applications are more uniform than are data applications, and
406 therefore the commercial services more often have what the voice
407 customer wants at a price that is competitive with the private
408 network option. Data communication requirements are still more
409 specialized and dynamic. Thus, there is less opportunity for economy
410 of scale in service offerings and it is harder to keep up to date
411 with customer demand. For this reason we expect private data networks
412 to persist for the near future. As the telephone companies begin to
413 introduce the next generation of high speed packet switched services,
414 the scenario should change. However, we maintain that the result will
415 be a predominance, but not complete dominance, of public carrier use
416 for long haul communication. Therefore, private data networks will
417 persist and the routing architecture must accommodate controlled
418 interconnection.) Critical differences between the two models follow
419 from the difference in assumptions regarding AD topology. In the
420 complex case, lateral connections must be supported, along with the
421 means to control the use of such connections in the routing
422 protocols.
423
424 The different topologies imply different policy requirements. The
425 first model assumes that all policies can be expressed and enforced
426 in terms of dollars and cents and distributed charging schemes. The
427 second model assumes that ADs want more varied control over their
428 resources, control that can not be captured in a dollars and cents
429 metric alone. We describe the types of policies to be supported and
430 provide examples in the following section, <a href="#section-6">Section 6</a>. In brief, given
431 private lateral links, ADs must be able to express access and
432 charging related restrictions and privileges that discriminate on an
433 AD basis. These policies will be diverse, dynamic, and new
434 requirements will emerge over time, consequently support must be
435 extensible. For example, the packaging and charging schemes of any
436 single long haul service will vary over time and may be relatively
437 elaborate (e.g., many tiers of service, special package deals, to
438 achieve price discrimination).
439
440 Note that these assumptions about complexity do not preclude some
441
442
443
444 <span class="grey">Estrin [Page 8]</span></pre>
445 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-9" ></span>
446 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
447
448
449 collection of ADs from "negotiating away" their policy differences,
450 i.e., forming a federation, and coordinating a simplified inter-AD
451 configuration in order to reduce the requirements for inter-AD
452 mechanisms. However, we maintain that there will persist collections
453 of ADs that will not and can not behave as a single federation; both
454 in the research community and, even more predominantly, in the
455 broader commercial arena. Moreover, when it comes to interconnecting
456 across these federations, non-negotiable differences will arise
457 eventually. It is our goal to develop mechanisms that are applicable
458 in the broader arena.
459
460 The Internet community developed its original protocol suite with
461 only minimal provision for resource control [<a href="#ref-9" title="Palo Alto">9</a>]. This was
462 appropriate at the time of development based on the assumed community
463 (i.e., researchers) and the ground breaking nature of the technology.
464 The next generation of network technology is now being designed to
465 take advantage of high speed media and to support high demand traffic
466 generated by more powerful computers and their applications [<a href="#ref-10" title="B. Leiner">10</a>]. As
467 with TCP/IP we hope that the technology being developed will find
468 itself applied outside of the research community. This time it would
469 be inexcusable to ignore resource control requirements and not to pay
470 careful attention to their specification.
471
472 Finally, we look forward to the Internet structure taking advantage
473 of economies of scale offered by enhanced commercial services.
474 However, in many respects the problem that stub-ADs may thus avoid,
475 will be faced by the multiple regional and long haul carriers
476 providing the services. The carriers' charging and resource control
477 policies will be complex enough to require routing mechanisms similar
478 to ones being proposed for the complex AD topology case described
479 here. Whether the network structure is based on private or
480 commercial services, the goal is to construct policy sensitive
481 mechanisms that will be transparent to end users (i.e., the
482 mechanisms are part of the routing infrastructure at the network
483 level, and not an end to end concern).
484
485 <span class="h2"><a class="selflink" id="section-6" href="#section-6">6</a> POLICY TYPES</span>
486
487 This section outlines a taxonomy of internet policies for inter-AD
488 topologies that allow lateral and bypass links. The taxonomy is
489 intended to cover a wide range of ADs and internets. Any particular
490 PR architecture we design should support a significant subset of
491 these policy types but may not support all of them due to technical
492 complexity and performance considerations. The general taxonomy is
493 important input to a functional specification for PR. Moreover, it
494 can be used to evaluate and compare the suitability and completeness
495 of existing routing architectures and protocols for PR; see <a href="#section-8">Section</a>
496 <a href="#section-8">8</a>.
497
498
499
500 <span class="grey">Estrin [Page 9]</span></pre>
501 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-10" ></span>
502 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
503
504
505 We provide examples from the Research Internet of the different
506 policy types in the form of resource usage policy statements. These
507 statements were collected through interviews with agency
508 representatives, but they do not represent official policy. These
509 sample policy statements should not} be interpreted as agency policy,
510 they are provided here only as examples.
511
512 Internet policies fall into two classes, access and charging. Access
513 policies specify who can use resources and under what conditions.
514 Charging policies specify the metering, accounting, and billing
515 implemented by a particular AD.
516
517 <span class="h3"><a class="selflink" id="section-6.1" href="#section-6.1">6.1</a> TAXONOMY OF ACCESS POLICIES</span>
518
519 We have identified the following types of access policies that ADs
520 may wish to enforce. Charging policies are described in the
521 subsequent section. <a href="#section-6.3">Section 6.3</a> provides more specific examples of
522 both access and charging policies using FRICC policy statements.
523
524 Access policies typically are expressed in the form: principals of
525 type x can have access to resources of type y under the following
526 conditions, z. The policies are categorized below according to the
527 definition of y and z. In any particular instance, each of the
528 policy types would be further qualified by definition of legitimate
529 principals, , x, i.e., what characteristics x must have in order to
530 access the resource in question.
531
532 We refer to access policies described by stub and transit ADs. The
533 two roles imply different motivations for resource control, however
534 the types of policies expressed are similar; we expect the supporting
535 mechanisms to be common as well.
536
537 Stub and transit access policies may specify any of the following
538 parameters:
539
540 * SOURCE/DESTINATION
541 Source/Destination policies prevent or restrict communication
542 originated by or destined for particular ADs (or hosts or user
543 classes within an AD).
544
545 * PATH
546 Path sensitive policies specify which ADs may or may not be passed
547 through en route to a destination. The most general path sensitive
548 policies allow stub and transit ADs to express policies that depend
549 on any component in the AD path. In other words, a stub AD could
550 reject a route based on any AD (or combination of ADs) in the route.
551 Similarly, a transit AD could express a packet forwarding policy that
552 behaves differently depending upon which ADs a packet has passed
553
554
555
556 <span class="grey">Estrin [Page 10]</span></pre>
557 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-11" ></span>
558 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
559
560
561 through, and is going to pass through, en route to the destination.
562 Less ambitious (and perhaps more reasonable) path sensitive policies
563 might only discriminate according to the immediate neighbor ADs
564 through which the packet is traveling (i.e., a stub network could
565 reject a route based on the first transit AD in the route, and a
566 transit AD could express a packet forwarding policy that depends upon
567 the previous, and the subsequent, transit ADs in the route.)
568
569 * QUALITY/TYPE OF SERVICE(QOS OR TOS)
570 This type of policy restricts access to special resources or
571 services. For example, a special high throughput, low delay link may
572 be made available on a selective basis.
573
574 * RESOURCE GUARANTEE
575 These policies provide a guaranteed percentage of a resource on a
576 selective, as needed basis. In other words, the resource can be used
577 by others if the preferred-AD's offered load is below the guaranteed
578 level of service. The guarantee may be to always carry intra-AD
579 traffic or to always carry inter-AD traffic for a specific AD.
580
581 * TEMPORAL
582 Temporal policies restrict usage based on the time of day or other
583 time related parameters.
584
585 * HIGH LEVEL PROTOCOL
586 Usage may be restricted to a specific high level protocol such as
587 mail or file transfer. (Alternatively, such policies can be
588 implemented as source/destination policies by configuring a host(s)
589 within an AD as an application relay and composing policy terms that
590 allow inter-AD access to only that host.)
591
592 * RESOURCE LIMIT
593 There may be a limit on the amount of traffic load a source may
594 generate during a particular time interval, e.g., so many packets in
595 a day, hour, or minute.
596
597 * AUTHENTICATION REQUIREMENTS
598 Conditions may be specified regarding the authenticability of
599 principal identifying information. Some ADs might require some form
600 of cryptographic proof as to the identity and affiliations of the
601 principal before providing access to critical resources.
602
603 The above policy types usually exist in combination for a particular
604 AD, i.e., an AD's policies might express a combination of transit,
605 source/destination, and QOS restrictions. This taxonomy will evolve
606 as PR is applied to other domains.
607
608 As will be seen in <a href="#section-6.3">Section 6.3</a> an AD can express its charging and
609
610
611
612 <span class="grey">Estrin [Page 11]</span></pre>
613 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-12" ></span>
614 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
615
616
617 access policies in a single syntax. Moreover, both stub and transit
618 policies can co-exist. This is important since some ADs operate as
619 both stub and transit facilities and require such hybrid control.
620
621 <span class="h3"><a class="selflink" id="section-6.2" href="#section-6.2">6.2</a> TAXONOMY OF CHARGING POLICIES</span>
622
623 Stub and transit charging policies may specify the following
624 parameters:
625
626 * UNIT OF ACCOUNTING (e.g., dollars or credits).
627 * BASIS FOR CHARGING (e.g., per Kbyte or per Kpkt).
628 * ACTUAL CHARGES (e.g., actual numbers such as $.50/Mbyte).
629 * WHO IS CHARGED OR PAID (e.g., originator of packet,
630 immediate neighbor from whom packet was received, destination
631 of packet, a third party collection agent).
632 * WHOSE PACKET COUNT is used (e.g., source, destination, the
633 transit AD's own count, the count of some upstream or
634 downstream AD).
635 * BOUND ON CHARGES (e.g., to limit the amount that a stub
636 AD is willing to spend, or the amount that a transit AD is
637 willing to carry.)
638
639 The enforcement of these policies may be carried out during route
640 synthesis or route selection [<a href="#ref-4" title="Policy Routing in Internet Protocols">4</a>].
641
642 <span class="h3"><a class="selflink" id="section-6.3" href="#section-6.3">6.3</a> EXAMPLE POLICY STATEMENTS</span>
643
644 The following policy statements were collected in the fall of 1988
645 through interviews with representatives of the federal agencies most
646 involved in supporting internetworking. Once again we emphasize that
647 these are not official policy statements. They are presented here to
648 provide concrete examples of the sort of policies that agencies would
649 like to enforce.
650
651 Expressing policies as Policy Terms (PTs)
652
653 Each policy is described in English and then expressed in a policy
654 term (PT) notation suggested by Dave Clark in [<a href="#ref-4" title="Policy Routing in Internet Protocols">4</a>]. Each PT
655 represents a distinct policy of the AD that synthesized it. The
656 format of a PT is:
657
658 [(H{src},AD{src},AD{ent}),(H{dst},AD{dst},AD{exit}),UCI, Cg,Cb]
659
660 Hsrc stands for source host, ADsrc for source AD, ADent for entering
661 AD (i.e., neighboring AD from which traffic is arriving directly),
662 Hdst for destination host, ADdst for destination AD, ADexit for exit
663 AD (i.e.,neighboring AD to which traffic is going directly), UCI for
664 user class identifier, and Cg and Cb for global and bilateral
665
666
667
668 <span class="grey">Estrin [Page 12]</span></pre>
669 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-13" ></span>
670 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
671
672
673 conditions, respectively. The purpose of a PT is to specify that
674 packets from some host, H{src}, (or a group of hosts) in a source AD,
675 AD{src}, are allowed to enter the AD in question via some directly
676 connected AD, AD{ent}, and exit through another directly connected
677 AD, AD{exit}, on its way to a host, H{dst}, (or a group of hosts) in
678 some destination AD, AD{dst}. User Class Identifier (UCI) allows for
679 distinguishing between various user classes, e.g., Government,
680 Research, Commercial, Contract, etc. Global Conditions (Cg)
681 represent billing and other variables. Bilateral Conditions (Cb)
682 relate to agreements between neighboring ADs, e.g., related to
683 metering or charging. In the example policy terms provided below we
684 make use of the following abbreviations: Fricc for
685 {DOE,NASA,DCA,NSF}, F for Federal Agency, Re for Regional, U for
686 University, Co for Commercial Corporation, and Cc for Commercial
687 Carrier. A hyphen, -, means no applicable matches.
688
689 By examining a PT we can identify the type of policy represented, as
690 per the taxonomy presented earlier.
691
692 * If an AD specifies a policy term that has a null (-) entry for
693 the ADexit, then it is disallowing transit for some group of users,
694 and it is a transit policy.
695
696 * If an AD specifies a policy term that lists itself
697 explicitly as ADsrc or ADdst, it is expressing restrictions on who
698 can access particular resources within its boundaries, or on who inside
699 can obtain external access. In other words the AD is expressing a
700 source/destination policy.
701
702 * If ADexit or ADentr is specified then the policy expressed is an
703 exit/entrance path policy.
704
705 * If the global conditions include charging, QOS, resource
706 guarantee, time of day, higher level application, resource limit, or
707 authentication related information it is obviously a charging, QOS,
708 resource guarantee, temporal, higher level application, resource
709 limit, or authentication policy, respectively.
710
711 As seen below, any one PT typically incorporates a combination of
712 policy types.
713
714 <span class="h4"><a class="selflink" id="section-6.3.1" href="#section-6.3.1">6.3.1</a> THE FRICC</span>
715
716 In the following examples all policies (and PTs) are symmetrical
717 under the assumption that communication is symmetrical.
718
719
720
721
722
723
724 <span class="grey">Estrin [Page 13]</span></pre>
725 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-14" ></span>
726 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
727
728
729 NATIONAL SCIENCE FOUNDATION (NSF)
730
731 1. NSF will carry traffic for any host connected to a F/Re network
732 talking to any other host connected to a F/Re via any F/Re entry and
733 exit network, so long as there is it is being used for research or
734 support. There is no authentication of the UCI and no per packet
735 charging. NSFnet is a backbone and so does not connect directly to
736 universities or companies...thus the indication of {F/Re} instead of
737 {F/Re/U/Co} as ADent and ADexit.
738
739 [NSF1: (*, {F/Re}, {F/Re})(*, {F/Re}, {F/Re}){research,support}
740 {unauthenticated UCI,no-per-pkt charge}{}]
741
742 2. NSF will carry traffic to user and expert services hosts in NSF
743 AD to/from any F/Re AD, via any F/Re AD. These are the only things
744 that directly connect to NSFnet.
745
746 [NSF2: ({User svcs, Expert Svcs},{NSF},{F/Re})(*,{F/Re},{-}){}{}{}]
747
748 DEPARTMENT OF ENERGY (DOE)
749
750 1. DOE will carry traffic to and from any host directly connected to
751 DOE so long as it is used for research or support. There is no
752 authentication of the UCI and no per packet charging.
753
754 [DOE1: (*,DOE,-)(*,*,*){research,support}
755 {unauthenticated UCI,no-per-packet charge}{}]
756
757 2. DOE will carry traffic for any host connected to a F/Re network
758 talking to any other host connected to a F/Re via any F/Re entry and
759 exit network without regard to the UCI. There is no authentication of
760 the UCI and no per packet charging. (in other words DOE is more
761 restrictive with its own traffic than with traffic it is carrying as
762 part of a resource sharing arrangement.)
763
764 [DOE2: (*,{F/Re},{F/Re})(*,{F/Re},{F/Re}){}
765 {unauthenticated UCI, no-per-pkt charge}{}]
766
767 NATIONAL AERONAUTICS AND SPACE ADMINISTRATION (NASA)
768
769 1. Nasa will accept any traffic to/from members of the Nasa AD. But
770 no transit. No UCI authentication and no per packet charge.
771
772 [NASA1: (*,*,*)(*,Nasa,-){Nasa-research, support}
773 {unauthenticated UCI,no-per-packet-charge}{}]
774
775 2. Nasa will carry transit traffic to/from other federal agency
776 networks if it is in support of research, and if the total use of
777
778
779
780 <span class="grey">Estrin [Page 14]</span></pre>
781 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-15" ></span>
782 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
783
784
785 available BW by non-nasa Federal agencies is below n%. NOTE THAT this
786 non-interference policy type needs some more work in terms of
787 integrating it into the routing algorithms. See <a href="#section-7">Section 7</a>.
788
789 [NASA2: (*,{F},*)(*,{F},*){research,support}
790 {per-packet accounting, limited to n% of available BW}{}]
791
792 3. NASA will carry commercial traffic to federal and regional and
793 university ADs for nasa research or support. But it will not allow
794 transit. The particular entry AD is not important.
795
796 [NASA3: (*,{Co},*} (*,{F/R/U},*) {NASA research,support}
797 {unauthenticated UCI, no per packet charge}{}]
798
799 4. On a case by case basis NASA may provide access to its resources
800 on a cost reimbursed basis. Transit traffic will not be carried on
801 this basis.
802
803 [NASA4: (*,*,-)(*,*,-){}
804 {per-packet-charge, limited to n% of available BW} {}]
805
806 DEFENSE ADVANCED RESEARCH PROJECTS AGENCY (DARPA)
807
808 1. DARPA will carry traffic to/from any host in DARPA AD from any
809 external host that can get it there so long as UCI is research or
810 support. No UCI authentication or per packet charge.
811
812 [DARPA1: (*,*,*)(*,DARPA,-){research,support}
813 {unauthenticated-UCI, no per packet charge}{}]
814
815 2. DARPA will carry traffic for any host connected to a F/Re/U/Co
816 network talking to any other host connected to a F/Re/U/Co via any
817 F/Re/U/Co entry and exit network, so long as there is it is being
818 used for research or support, and the network is not heavily
819 congested!!. There is no authentication of the UCI and no per packet
820 charging. NOTE: Darpa would like to say something about the need to
821 enter the Darpa AD at the point closest to the destination...but i
822 don't know how to express this...
823
824 DARPA2: (*,{F/R/U/Co},{F/R/U/Co})(*,{F/R/U/Co},{F/R/U/Co})
825 {research,support}{unauthenticated-UCI,no per packet charge,
826 non-interference basis}{}]
827
828 DEFENSE COMMUNICATIONS AGENCY (DCA)
829
830 1. DCA will not carry any transit traffic. It will only accept and
831 send traffic to and from its mailbridge(s) and only from and to hosts
832 on other F/Re nets. All packets are marked and charged for by the
833
834
835
836 <span class="grey">Estrin [Page 15]</span></pre>
837 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-16" ></span>
838 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
839
840
841 kilopacket.
842
843 [DCA1:(mailbridge,DCA,-)(*,{F/Re},{F/Re}){research,support}
844 {unauthenticated UCI, all incoming packets marked, per-kilopacket
845 charge}{}]
846
847 <span class="h4"><a class="selflink" id="section-6.3.2" href="#section-6.3.2">6.3.2</a> THE REGIONALS</span>
848
849 Interviews with regional network administrations are now underway. In
850 general their policies are still in formation due to the relatively
851 recent formation of these regional networks. However, for the sake of
852 illustration we provide an example of a hypothetical regional's
853 network policies.
854
855 REGIONAL A
856
857 1. Regional A will carry traffic from/to any directly connected
858 F/Re/U network to any F/Re/U network via NSF if it is for a research
859 or support UCI. (NSF requires that all Regional networks only pass it
860 traffic that complies with its, NSF's, policies!)
861
862 [Regional A:(*,{F/Re/U},{F/Re/U})(*,{F/Re/U},NSF){research,support}
863 {unauthenticated UCI, no-per-packet charge}{}]
864
865 REGIONAL B
866
867 1. Regional B will carry traffic from/to any directly connected
868 F/Re/U network to any F/Re/U network via a commercial carrier
869 regardless of its UCI. In this case the packets are charged for since
870 the commercial carrier charges per kilopacket.
871
872 [Regional B:(*,{F/Re/U},{F/Re/U})(*,{F/Re/U},Cc){}
873 {unauthenticated UCI, per-kilopacket charge}{}]
874
875 <span class="h4"><a class="selflink" id="section-6.3.3" href="#section-6.3.3">6.3.3</a> CAMPUS AND PRIVATE NETWORKS</span>
876
877 Similar interviews should be conducted with administrators of campus
878 and private networks. However, many aspects of their policies are
879 contingent on the still unresolved policies of the regionals and
880 federal agencies. In any event, transit policies will be critical
881 for campus and private networks to flexibly control access to lateral
882 links and private wide area networks, respectively. For example, a
883 small set of university and private laboratories may provide access
884 to special gigabit links for particular classes of researchers. On
885 the other hand, source/destination policies should not be used in
886 place of network level access controls for these end ADs.
887
888
889
890
891
892 <span class="grey">Estrin [Page 16]</span></pre>
893 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-17" ></span>
894 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
895
896
897 <span class="h4"><a class="selflink" id="section-6.3.4" href="#section-6.3.4">6.3.4</a> COMMERCIAL SERVICES</span>
898
899 Currently commercial communication services play a low level role in
900 most parts of today's Research Internet; they provide the
901 transmission media, i.e.,leased lines. In the future we expect
902 commercial carriers to provide increasingly higher level and enhanced
903 services such as high speed packet switched backbone services.
904 Because such services are not yet part of the Research Internet
905 infrastructure there exist no policy statements.
906
907 Charging and accounting are certain to be an important policy type in
908 this context. Moreover, we anticipate the long haul services market
909 to be highly competitive. This implies that competing service
910 providers will engage in significant gaming in terms of packaging and
911 pricing of services. Consequently, the ability to express varied and
912 dynamic charging policies will be critical for these ADs.
913
914 <span class="h2"><a class="selflink" id="section-7" href="#section-7">7</a> PROBLEMATIC REQUIREMENTS</span>
915
916 Most of this paper has lobbied for articulation of relatively
917 detailed policy statements in order to help define the technical
918 mechanisms needed for enforcement. We promoted a top down design
919 process beginning with articulation of desired policies. Now we feel
920 compelled to mention requirements that are clearly problematic from
921 the bottom up perspective of technical feasibility.
922
923 * Non-interference policies are of the form "I will provide
924 access for principals x to resources y so long as it does not
925 interfere with my internal usage." The problem with such policies
926 is that access to an AD at any point in time is contingent upon a
927 local, highly dynamic, parameter that is not globally available.
928 Therefore such a policy term could well result in looping,
929 oscillations, and excessive route (re)computation overhead,
930 both unacceptable. Consequently, this is one type of policy that
931 routing experts suggest would be difficult to support in a very
932 large decentralized internetwork.
933
934 * Granularity can also be problematic, but not as devistating as
935 highly dynamic PR contingencies. Here the caution is less specific.
936 Very fine grain policies, which restrict access to particular
937 hosts, or are contingent upon very fine grain user class
938 identification, may be achieved more efficiently with network
939 level access control [<a href="#ref-11" title="J. Mogul and G. Tsudik">11</a>] or end system controls instead of
940 burdening the inter-AD routing mechanism.
941
942 * Security is expensive, as always. Routing protocols are subject
943 to fraud through impersonation, data substitution, and denial of
944 service. Some of the proposed mechanisms provide some means for
945
946
947
948 <span class="grey">Estrin [Page 17]</span></pre>
949 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-18" ></span>
950 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
951
952
953 detection and non-repudiation. However, to achieve a priori
954 prevention of resource misuse is expensive in terms of per
955 connection or per packet cryptographic overhead. For some
956 environments we firmly believe that this will be necessary and
957 we would prefer an architecture that would accommodate such
958 variability [<a href="#ref-12" title="Security Issues in Policy Routing">12</a>].
959
960 In general, it is difficult to predict the impact of any particular
961 policy term. Tools will be needed to assist people in writing and
962 validating policy terms.
963
964 <span class="h2"><a class="selflink" id="section-8" href="#section-8">8</a> PROPOSED MECHANISMS</span>
965
966 Previous routing protocols have addressed a narrower definition of
967 PR, as appropriate for the internets of their day. In particular, EGP
968 [<a href="#ref-3" title="Exterior Gateway Protocol (EGP)">3</a>], DGP[13], and BGP[6] incorporate a notion of policy restrictions
969 as to where routing database information travels. None are intended
970 to support policy based routing of packets as described here. More
971 recent routing proposals such as Landmark [<a href="#ref-14" title="Palo Alto">14</a>] and Cartesian [<a href="#ref-15" title="Technical Report">15</a>]
972 could be used to restrict packet forwarding but are not suited to
973 source/destination, and some of the condition-oriented, policies. We
974 feel these policy types are critical to support. We note that for
975 environments (e.g., within an AD substructure) in which the simple-
976 AD-topology conjecture holds true, these alternatives may be
977 suitable.
978
979 <a href="./rfc1104">RFC 1104</a> [<a href="#ref-5" title="Models of Policy Based Routing">5</a>] provides a good description of shorter term policy
980 routing requirements. Braun classifies three types of mechanisms,
981 policy based distribution of route information, policy based packet
982 forwarding, and policy based dynamic allocation of network resources.
983 The second class is characterized by Dave Clark's PR architecture,
984 <a href="./rfc1102">RFC 1102</a> [<a href="#ref-4" title="Policy Routing in Internet Protocols">4</a>]. With respect to the longer term requirements laid out
985 in this document, only this second class is expressive and flexible
986 enough to support the multiplicity of stub and transit policies. In
987 other words, the power of the PR approach (e.g., <a href="./rfc1102">RFC1102</a>) is not just
988 in the added granularity of control pointed out by Braun, i.e., the
989 ability to specify particular hosts and user classes. Its power is in
990 the ability to express and enforce many types of stub and transit
991 policies and apply them on a discriminatory basis to different ADs.
992 In addition, this approach provides explicit support for stub ADs to
993 control routes via the use of source routing. (FOOTNOTE 12:
994 Moreover, the source routing approach loosens the requirements for
995 every AD to share a complete view of the entire internet by allowing
996 the source to detect routing loops.) (FOOTNOTE 13: The match
997 between <a href="./rfc1102">RFC1102</a> and the requirements specified in this document is
998 hardly a coincidence since Clark's paper and discussions with him
999 contributed to the requirements formulation presented here. His work
1000 is currently being evaluated and refined by the ANRG and ORWG.)
1001
1002
1003
1004 <span class="grey">Estrin [Page 18]</span></pre>
1005 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-19" ></span>
1006 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
1007
1008
1009 <span class="h2"><a class="selflink" id="section-9" href="#section-9">9</a> SUMMARY</span>
1010
1011 Along with the emergence of very high speed applications and media,
1012 resource management has become a critical issue in the Research
1013 Internet and internets in general. A fundamental characteristic of
1014 the resource management problem is allowing administratively ADs to
1015 interconnect while retaining control over resource usage. However, we
1016 have lacked a careful articulation of the types of resource
1017 management policies that need to be supported. This paper addresses
1018 policy requirements for the Research Internet. After justifying our
1019 assumptions regarding AD topology we presented a taxonomy and
1020 examples of policies that must be supported by a PR protocol.
1021
1022 <span class="h2"><a class="selflink" id="section-10" href="#section-10">10</a> ACKNOWLEDGMENTS</span>
1023
1024 Members of the Autonomous Networks Research Group and Open Routing
1025 Working Group have contributed significantly to the ideas presented
1026 here, in particular, Guy Almes, Lee Breslau, Scott Brim, Dave Clark,
1027 Marianne Lepp, and Gene Tsudik. In addition, Lee Breslau and Gene
1028 Tsudik provided detailed comments on a previous draft. David Cheriton
1029 inadvertently caused me to write this document. Sharon Anderson's
1030 contributions deserve special recognition. The author is supported
1031 by research grants from National Science Foundation, AT&amp;T, and GTE.
1032
1033 <span class="h2"><a class="selflink" id="section-11" href="#section-11">11</a> REFERENCES</span>
1034
1035 [<a id="ref-1">1</a>] J. Postel, Internet Protocol, Network Information Center, <a href="./rfc791">RFC</a>
1036 <a href="./rfc791">791</a>, September 1981.
1037
1038 [<a id="ref-2">2</a>] G. Vaudreuil, The Federal Research Internet Coordinating
1039 Committee and National Research Network, ACM SIG Computer
1040 Communications Review,April 1988.
1041
1042 [<a id="ref-3">3</a>] E. Rosen, Exterior Gateway Protocol (EGP), Network Information
1043 Center, <a href="./rfc827">RFC 827</a>, October 1982.
1044
1045 [<a id="ref-4">4</a>] D. Clark, Policy Routing in Internet Protocols, Network
1046 Information Center, <a href="./rfc1102">RFC 1102</a>, May 1989.
1047
1048 [<a id="ref-5">5</a>] H.W.Braun, Models of Policy Based Routing, Network Information
1049 Center, <a href="./rfc1104">RFC 1104</a>, June 1989.
1050
1051 [<a id="ref-6">6</a>] K. Lougheed, Y. Rekhter, A Border Gateway Protocol, Network
1052 Information Center, <a href="./rfc1105">RFC 1105</a>, June 1989.
1053
1054 [<a id="ref-7">7</a>] J. Saltzer, M. Schroeder, The Protection of Information in
1055 Computer Systems, Proceedings of the IEEE, 63, 9 September 1975.
1056
1057
1058
1059
1060 <span class="grey">Estrin [Page 19]</span></pre>
1061 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-20" ></span>
1062 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
1063
1064
1065 [<a id="ref-8">8</a>] V. Jacobson, Congestion Avoidance and Control. Proceedings of
1066 ACM Sigcomm, pp. 106-114, August 1988, Palo Alto, CA.
1067
1068 [<a id="ref-9">9</a>] David Clark, Design Philosophy of the DARPA Internet Protocols,
1069 Proceedings of ACM Sigcomm, pp. 106-114, August 1988, Palo Alto,
1070 CA.
1071
1072 [<a id="ref-10">10</a>] Gigabit Networking Group, B. Leiner, Editor. Critical Issues in
1073 High Bandwidth Networking, Network Information Center, <a href="./rfc1077">RFC 1077</a>,
1074 November 1988.
1075
1076 [<a id="ref-11">11</a>] D. Estrin, J. Mogul and G. Tsudik, Visa Protocols for Controlling
1077 Inter-Organizational Datagram Flow, To appear in IEEE Journal on
1078 Selected Areas in Communications, Spring 1989.
1079
1080 [<a id="ref-12">12</a>] D. Estrin and G. Tsudik, Security Issues in Policy Routing, IEEE
1081 Symposium on Research in Security and Privacy, Oakland, CA. May
1082 1-3 1989.
1083
1084 [<a id="ref-13">13</a>] M. Little, The Dissimilar Gateway Protocol, Technical report
1085
1086 [<a id="ref-14">14</a>] P. Tsuchiya, The Landmark Hierarchy: A new hierarchy for routing
1087 in very large networks, IEEE SIGCOMM 88, Palo Alto, CA. September
1088 1988.
1089
1090 [<a id="ref-15">15</a>] G. Finn, Reducing the Vulnerability of Dynamic Computer Networks
1091 USC/Information Sciences Institute, Technical Report, ISI/RR-88-
1092 201 July 1988.
1093
1094 [<a id="ref-16">16</a>] A. Nakassis Routing Algorithm for Open Routing, Unpublished
1095 paper, Available from the author at the National Institute of
1096 Standards and Technology (formerly NBS), Washington D.C.
1097
1098 <span class="h2"><a class="selflink" id="section-11" href="#section-11">11</a> SECURITY CONSIDERATIONS</span>
1099
1100 This memo does not address the security aspects of the issues
1101 discussed.
1102
1103 AUTHOR'S ADDRESS:
1104
1105 Deborah Estrin
1106 University of Southern California
1107 Computer Science Department
1108 Los Angeles, CA 90089-0782
1109
1110 Phone: (213) 743-7842
1111
1112 EMail: [email protected]
1113
1114
1115
1116 <span class="grey">Estrin [Page 20]</span></pre>
1117 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-21" ></span>
1118 <span class="grey"><a href="./rfc1125">RFC 1125</a> Policy Requirements November 1989</span>
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172 Estrin [Page 21]
1173 </pre>