Debian Codegraph @

rfc3539.html

raw
1 <pre>Network Working Group B. Aboba
2 Request for Comments: 3539 Microsoft
3 Category: Standards Track J. Wood
4 Sun Microsystems, Inc.
5 June 2003
6
7
8 <span class="h1">Authentication, Authorization and Accounting (AAA) Transport Profile</span>
9
10 Status of this Memo
11
12 This document specifies an Internet standards track protocol for the
13 Internet community, and requests discussion and suggestions for
14 improvements. Please refer to the current edition of the "Internet
15 Official Protocol Standards" (STD 1) for the standardization state
16 and status of this protocol. Distribution of this memo is unlimited.
17
18 Copyright Notice
19
20 Copyright (C) The Internet Society (2003). All Rights Reserved.
21
22 Abstract
23
24 This document discusses transport issues that arise within protocols
25 for Authentication, Authorization and Accounting (AAA). It also
26 provides recommendations on the use of transport by AAA protocols.
27 This includes usage of standards-track RFCs as well as experimental
28 proposals.
29
30 Table of Contents
31
32 <a href="#section-1">1</a>. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . <a href="#page-2">2</a>
33 <a href="#section-1.1">1.1</a>. Requirements Language. . . . . . . . . . . . . . . . . . <a href="#page-2">2</a>
34 <a href="#section-1.2">1.2</a>. Terminology. . . . . . . . . . . . . . . . . . . . . . . <a href="#page-2">2</a>
35 <a href="#section-2">2</a>. Issues in Transport Usage. . . . . . . . . . . . . . . . . . . <a href="#page-5">5</a>
36 <a href="#section-2.1">2.1</a>. Application-driven Versus Network-driven . . . . . . . . <a href="#page-5">5</a>
37 <a href="#section-2.2">2.2</a>. Slow Failover. . . . . . . . . . . . . . . . . . . . . . <a href="#page-6">6</a>
38 <a href="#section-2.3">2.3</a>. Use of Nagle Algorithm . . . . . . . . . . . . . . . . . <a href="#page-7">7</a>
39 <a href="#section-2.4">2.4</a>. Multiple Connections . . . . . . . . . . . . . . . . . . <a href="#page-7">7</a>
40 <a href="#section-2.5">2.5</a>. Duplicate Detection. . . . . . . . . . . . . . . . . . . <a href="#page-8">8</a>
41 <a href="#section-2.6">2.6</a>. Invalidation of Transport Parameter Estimates. . . . . . <a href="#page-8">8</a>
42 <a href="#section-2.7">2.7</a>. Inability to use Fast Re-Transmit. . . . . . . . . . . . <a href="#page-9">9</a>
43 <a href="#section-2.8">2.8</a>. Congestion Avoidance . . . . . . . . . . . . . . . . . . <a href="#page-9">9</a>
44 <a href="#section-2.9">2.9</a>. Delayed Acknowledgments. . . . . . . . . . . . . . . . . <a href="#page-11">11</a>
45 <a href="#section-2.10">2.10</a>. Premature Failover . . . . . . . . . . . . . . . . . . . <a href="#page-11">11</a>
46 <a href="#section-2.11">2.11</a>. Head of Line Blocking. . . . . . . . . . . . . . . . . . <a href="#page-11">11</a>
47 <a href="#section-2.12">2.12</a>. Connection Load Balancing. . . . . . . . . . . . . . . . <a href="#page-12">12</a>
48
49
50
51
52 <span class="grey">Aboba &amp; Wood Standards Track [Page 1]</span></pre>
53 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-2" ></span>
54 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
55
56
57 <a href="#section-3">3</a>. AAA Transport Profile. . . . . . . . . . . . . . . . . . . . . <a href="#page-12">12</a>
58 <a href="#section-3.1">3.1</a>. Transport Mappings . . . . . . . . . . . . . . . . . . . <a href="#page-12">12</a>
59 <a href="#section-3.2">3.2</a>. Use of Nagle Algorithm . . . . . . . . . . . . . . . . . <a href="#page-12">12</a>
60 <a href="#section-3.3">3.3</a>. Multiple Connections . . . . . . . . . . . . . . . . . . <a href="#page-13">13</a>
61 <a href="#section-3.4">3.4</a>. Application Layer Watchdog . . . . . . . . . . . . . . . <a href="#page-13">13</a>
62 <a href="#section-3.5">3.5</a>. Duplicate Detection. . . . . . . . . . . . . . . . . . . <a href="#page-19">19</a>
63 <a href="#section-3.6">3.6</a>. Invalidation of Transport Parameter Estimates. . . . . . <a href="#page-20">20</a>
64 <a href="#section-3.7">3.7</a>. Inability to use Fast Re-Transmit. . . . . . . . . . . . <a href="#page-21">21</a>
65 <a href="#section-3.8">3.8</a>. Head of Line Blocking. . . . . . . . . . . . . . . . . . <a href="#page-22">22</a>
66 <a href="#section-3.9">3.9</a>. Congestion Avoidance . . . . . . . . . . . . . . . . . . <a href="#page-23">23</a>
67 <a href="#section-3.10">3.10</a>. Premature Failover . . . . . . . . . . . . . . . . . . . <a href="#page-24">24</a>
68 <a href="#section-4">4</a>. Security Considerations. . . . . . . . . . . . . . . . . . . . <a href="#page-24">24</a>
69 <a href="#section-5">5</a>. IANA Considerations. . . . . . . . . . . . . . . . . . . . . . <a href="#page-25">25</a>
70 <a href="#section-6">6</a>. References . . . . . . . . . . . . . . . . . . . . . . . . . . <a href="#page-25">25</a>
71 <a href="#section-6.1">6.1</a>. Normative References . . . . . . . . . . . . . . . . . . <a href="#page-25">25</a>
72 <a href="#section-6.2">6.2</a>. Informative References . . . . . . . . . . . . . . . . . <a href="#page-26">26</a>
73 <a href="#appendix-A">Appendix A</a> - Detailed Watchdog Algorithm Description . . . . . . . <a href="#page-28">28</a>
74 <a href="#appendix-B">Appendix B</a> - AAA Agents. . . . . . . . . . . . . . . . . . . . . . <a href="#page-33">33</a>
75 <a href="#appendix-B.1">B.1</a>. Relays and Proxies . . . . . . . . . . . . . . . . . . . <a href="#page-33">33</a>
76 <a href="#appendix-B.2">B.2</a>. Re-directs . . . . . . . . . . . . . . . . . . . . . . . <a href="#page-35">35</a>
77 <a href="#appendix-B.3">B.3</a>. Store and Forward Proxies. . . . . . . . . . . . . . . . <a href="#page-36">36</a>
78 <a href="#appendix-B.4">B.4</a>. Transport Layer Proxies. . . . . . . . . . . . . . . . . <a href="#page-38">38</a>
79 Intellectual Property Statement. . . . . . . . . . . . . . . . . . <a href="#page-39">39</a>
80 Acknowledgments. . . . . . . . . . . . . . . . . . . . . . . . . . <a href="#page-39">39</a>
81 Author Addresses . . . . . . . . . . . . . . . . . . . . . . . . . <a href="#page-40">40</a>
82 Full Copyright Statement . . . . . . . . . . . . . . . . . . . . . <a href="#page-41">41</a>
83
84 <span class="h2"><a class="selflink" id="section-1" href="#section-1">1</a>. Introduction</span>
85
86 This document discusses transport issues that arise within protocols
87 for Authentication, Authorization and Accounting (AAA). It also
88 provides recommendations on the use of transport by AAA protocols.
89 This includes usage of standards-track RFCs as well as experimental
90 proposals.
91
92 <span class="h3"><a class="selflink" id="section-1.1" href="#section-1.1">1.1</a>. Requirements Language</span>
93
94 In this document, the key words "MAY", "MUST, "MUST NOT", "optional",
95 "recommended", "SHOULD", and "SHOULD NOT", are to be interpreted as
96 described in [<a href="./rfc2119" title="&quot;Key words for use in RFCs to Indicate Requirement Levels&quot;">RFC2119</a>].
97
98 <span class="h3"><a class="selflink" id="section-1.2" href="#section-1.2">1.2</a>. Terminology</span>
99
100 Accounting
101 The act of collecting information on resource usage for the
102 purpose of trend analysis, auditing, billing, or cost
103 allocation.
104
105
106
107
108 <span class="grey">Aboba &amp; Wood Standards Track [Page 2]</span></pre>
109 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-3" ></span>
110 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
111
112
113 Administrative Domain
114 An internet, or a collection of networks, computers, and
115 databases under a common administration.
116
117 Agent A AAA agent is an intermediary that communicates with AAA
118 clients and servers. Several types of AAA agents exist,
119 including Relays, Re-directs, and Proxies.
120
121 Application-driven transport
122 Transport behavior is said to be "application-driven" when
123 the rate at which messages are sent is limited by the rate
124 at which the application generates data, rather than by the
125 size of the congestion window. In the most extreme case,
126 the time between transactions exceeds the round-trip time
127 between sender and receiver, implying that the application
128 operates with an effective congestion window of one. AAA
129 transport is typically application driven.
130
131 Attribute Value Pair (AVP)
132 The variable length concatenation of a unique Attribute
133 (represented by an integer) and a Value containing the
134 actual value identified by the attribute.
135
136 Authentication
137 The act of verifying a claimed identity, in the form of a
138 pre-existing label from a mutually known name space, as the
139 originator of a message (message authentication) or as the
140 end-point of a channel (entity authentication).
141
142 Authorization
143 The act of determining if a particular right, such as
144 access to some resource, can be granted to the presenter of
145 a particular credential.
146
147 Billing The act of preparing an invoice.
148
149 Network Access Identifier
150 The Network Access Identifier (NAI) is the userID submitted
151 by the host during network access authentication. In
152 roaming, the purpose of the NAI is to identify the user as
153 well as to assist in the routing of the authentication
154 request. The NAI may not necessarily be the same as the
155 user's e-mail address or the user-ID submitted in an
156 application layer authentication.
157
158
159
160
161
162
163
164 <span class="grey">Aboba &amp; Wood Standards Track [Page 3]</span></pre>
165 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-4" ></span>
166 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
167
168
169 Network Access Server (NAS)
170 A Network Access Server (NAS) is a device that hosts
171 connect to in order to get access to the network.
172
173 Proxy In addition to forwarding requests and responses, proxies
174 enforce policies relating to resource usage and
175 provisioning. This is typically accomplished by tracking
176 the state of NAS devices. While proxies typically do not
177 respond to client Requests prior to receiving a Response
178 from the server, they may originate Reject messages in
179 cases where policies are violated. As a result, proxies
180 need to understand the semantics of the messages passing
181 through them, and may not support all extensions.
182
183 Local Proxy
184 A Local Proxy is a proxy that exists within the same
185 administrative domain as the network device (e.g. NAS) that
186 issued the AAA request. Typically a local proxy is used to
187 multiplex AAA messages to and from a large number of
188 network devices, and may implement policy.
189
190 Store and forward proxy
191 Store and forward proxies distinguish themselves from other
192 proxy species by sending a reply to the NAS prior to
193 proxying the request to the server. As a result, store and
194 forward proxies need to implement AAA client and server
195 functionality for the messages that they handle. Store and
196 Forward proxies also typically keep state on conversations
197 in progress in order to assure delivery of proxied Requests
198 and Responses. While store and forward proxies are most
199 frequently deployed for accounting, they also can be used
200 to implement authentication/authorization policy.
201
202 Network-driven transport
203 Transport behavior is said to be "network driven" when the
204 rate at which messages are sent is limited by the
205 congestion window, not by the rate at which the application
206 can generate data. File transfer is an example of an
207 application where transport is network driven.
208
209 Re-direct Rather than forwarding Requests and Responses between
210 clients and servers, Re-directs refer clients to servers
211 and allow them to communicate directly. Since Re-directs
212 do not sit in the forwarding path, they do not alter any
213 AVPs transitting between client and server. Re-directs do
214 not originate messages and are capable of handling any
215 message type. A Re-direct may be configured only to re-
216 direct messages of certain types, while acting as a Relay
217
218
219
220 <span class="grey">Aboba &amp; Wood Standards Track [Page 4]</span></pre>
221 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-5" ></span>
222 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
223
224
225 or Proxy for other types. As with Relays, re-directs do
226 not keep state with respect to conversations or NAS
227 resources.
228
229 Relay Relays forward requests and responses based on routing-
230 related AVPs and domain forwarding table entries. Since
231 relays do not enforce policies, they do not examine or
232 alter non-routing AVPs. As a result, relays never
233 originate messages, do not need to understand the semantics
234 of messages or non-routing AVPs, and are capable of
235 handling any extension or message type. Since relays make
236 decisions based on information in routing AVPs and domain
237 forwarding tables they do not keep state on NAS resource
238 usage or conversations in progress.
239
240 <span class="h2"><a class="selflink" id="section-2" href="#section-2">2</a>. Issues in AAA Transport Usage</span>
241
242 Issues that arise in AAA transport usage include:
243
244 Application-driven versus network-driven
245 Slow failover
246 Use of Nagle Algorithm
247 Multiple connections
248 Duplicate detection
249 Invalidation of transport parameter estimates
250 Inability to use fast re-transmit
251 Congestion avoidance
252 Delayed acknowledgments
253 Premature Failover
254 Head of line blocking
255 Connection load balancing
256
257 We discuss each of these issues in turn.
258
259 <span class="h3"><a class="selflink" id="section-2.1" href="#section-2.1">2.1</a>. Application-driven versus Network-driven</span>
260
261 AAA transport behavior is typically application rather than network
262 driven. This means that the rate at which messages are sent is
263 typically limited by how quickly they are generated by the
264 application, rather than by the size of the congestion window.
265
266 For example, let us assume a 48-port NAS with an average session time
267 of 20 minutes. This device will, on average, send only 144
268 authentication/authorization requests/hour, and an equivalent number
269 of accounting requests. This represents an average inter-packet
270 spacing of 25 seconds, which is much larger than the Round Trip Time
271 (RTT) in most networks.
272
273
274
275
276 <span class="grey">Aboba &amp; Wood Standards Track [Page 5]</span></pre>
277 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-6" ></span>
278 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
279
280
281 Even on much larger NAS devices, the inter-packet spacing is often
282 larger than the RTT. For example, consider a 2048-port NAS with an
283 average session time of 10 minutes. It will on average send 3.4
284 authentication/authorization requests/second, and an equivalent
285 number of accounting requests. This translates to an average inter-
286 packet spacing of 293 ms.
287
288 However, even where transport behavior is largely application-driven,
289 periods of network-driven behavior can occur. For example, after a
290 NAS reboot, previously stored accounting records may be sent to the
291 accounting server in rapid succession. Similarly, after recovery
292 from a power failure, users may respond with a large number of
293 simultaneous logins. In both cases, AAA messages may be generated
294 more quickly than the network will allow them to be sent, and a queue
295 will build up.
296
297 Network congestion can occur when transport behavior is network-
298 driven or application-driven. For example, while a single NAS may
299 not send substantial AAA traffic, many NASes may communicate with a
300 single AAA proxy or server. As a result, routers close to a heavily
301 loaded proxy or server may experience congestion, even though traffic
302 from each individual NAS is light. Such "convergent congestion" can
303 result in dropped packets in routers near the AAA server, or even
304 within the AAA server itself.
305
306 Let us consider what happens when 10,000 48-ports NASes, each with an
307 average session time of 20 minutes, are configured with the same AAA
308 agent or server. The unfortunate proxy or server would receive 400
309 authentication/authorization requests/second and an equivalent number
310 of accounting requests. For 1000 octet requests, this would generate
311 6.4 Mbps of incoming traffic at the AAA agent or server.
312
313 While this transaction load is within the capabilities of the fastest
314 AAA agents and servers, implementations exist that cannot handle such
315 a high load. Thus high queuing delays and/or dropped packets may be
316 experienced at the agent or server, even if routers on the path are
317 not congested. Thus, a well designed AAA protocol needs to be able
318 to handle congestion occurring at the AAA server, as well as
319 congestion experienced within the network.
320
321 <span class="h3"><a class="selflink" id="section-2.2" href="#section-2.2">2.2</a>. Slow Failover</span>
322
323 Where TCP [<a href="./rfc793" title="&quot;Transmission Control Protocol&quot;">RFC793</a>] is used as the transport, AAA implementations will
324 experience very slow fail over times if they wait until a TCP
325 connection times out before resending on another connection. This is
326 not an issue for SCTP [<a href="./rfc2960" title="&quot;Stream Control Transmission Protocol&quot;">RFC2960</a>], which supports endpoint and path
327 failure detection. As described in <a href="./rfc2960#section-8">section&nbsp;8 of [RFC2960]</a>, when the
328 number of retransmissions exceeds the maximum
329
330
331
332 <span class="grey">Aboba &amp; Wood Standards Track [Page 6]</span></pre>
333 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-7" ></span>
334 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
335
336
337 ("Association.Max.Retrans"), the peer endpoint is considered
338 unreachable, the association enters the CLOSED state, and the failure
339 is reported to the application. This enables more rapid failure
340 detection.
341
342 <span class="h3"><a class="selflink" id="section-2.3" href="#section-2.3">2.3</a>. Use of Nagle Algorithm</span>
343
344 AAA protocol messages are often smaller than the maximum segment size
345 (MSS). While exceptions occur when certificate-based authentication
346 messages are issued or where a low path MTU is found, typically AAA
347 protocol messages are less than 1000 octets. Therefore, when using
348 TCP [<a href="./rfc793" title="&quot;Transmission Control Protocol&quot;">RFC793</a>], the total packet count and associated network overhead
349 can be reduced by combining multiple AAA messages within a single
350 packet.
351
352 Where AAA runs over TCP and transport behavior is network-driven,
353 such as after a reboot when many users login simultaneously, or many
354 stored accounting records need to be sent, the Nagle algorithm will
355 result in "transport layer batching" of AAA messages. While this
356 does not reduce the work required by the application in parsing
357 packets and responding to the messages, it does reduce the number of
358 packets processed by routers along the path. The Nagle algorithm is
359 not used with SCTP.
360
361 Where AAA transport is application-driven, the NAS will typically
362 receive a reply from the home server prior to having another request
363 to send. This implies, for example, that accounting requests will
364 typically be sent individually rather than being batched by the
365 transport layer. As a result, within the application-driven regime,
366 the Nagle algorithm [<a href="./rfc896" title="&quot;Congestion Control in IP/TCP internetworks&quot;">RFC896</a>] is ineffective.
367
368 <span class="h3"><a class="selflink" id="section-2.4" href="#section-2.4">2.4</a>. Multiple Connections</span>
369
370 Since the RADIUS [<a href="./rfc2865" title="&quot;Remote Authentication Dial In User Service (RADIUS)&quot;">RFC2865</a>] Identifier field is a single octet, a
371 maximum of 256 requests can be in progress between two endpoints
372 described by a 5-tuple: (Client IP address, Client port, UDP, Server
373 IP address, Server port). In order to get around this limitation,
374 RADIUS clients have utilized more than one sending port, sometimes
375 even going to the extreme of using a different UDP source port for
376 each NAS port.
377
378 Were this behavior to be extended to AAA protocols operating over
379 reliable transport, the result would be multiplication of the
380 effective slow-start ramp-up by the number of connections. For
381 example, if a AAA client had ten connections open to a AAA agent, and
382 used a per-connection initial window [<a href="./rfc3390" title="&quot;Increasing TCP's Initial Window&quot;">RFC3390</a>] of 2, then the
383
384
385
386
387
388 <span class="grey">Aboba &amp; Wood Standards Track [Page 7]</span></pre>
389 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-8" ></span>
390 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
391
392
393 effective initial window would be 20. This is inappropriate, since
394 it would permit the AAA client to send a large burst of packets into
395 the network.
396
397 <span class="h3"><a class="selflink" id="section-2.5" href="#section-2.5">2.5</a>. Duplicate Detection</span>
398
399 Where a AAA client maintains connections to multiple AAA agents or
400 servers, and where failover/failback or connection load balancing is
401 supported, it is possible for multiple agents or servers to receive
402 duplicate copies of the same transaction. A transaction may be sent
403 on another connection before expiration of the "time wait" interval
404 necessary to guarantee that all packets sent on the original
405 connection have left the network. Therefore it is conceivable that
406 transactions sent on the alternate connection will arrive before
407 those sent on the failed connection. As a result, AAA agents and
408 servers MUST be prepared to handle duplicates, and MUST assume that
409 duplicates can arrive on any connection.
410
411 For example, in billing, it is necessary to be able to weed out
412 duplicate accounting records, based on the accounting session-id,
413 event-timestamp and NAS identification information. Where
414 authentication requests are always idempotent, the resultant
415 duplicate responses from multiple servers will presumably be
416 identical, so that little harm will result.
417
418 However, there are situations where the response to an authentication
419 request will depend on a previously established state, such as when
420 simultaneous usage restrictions are being enforced. In such cases,
421 authentication requests will not be idempotent. For example, while
422 an initial request might elicit an Accept response, a duplicate
423 request might elicit a Reject response from another server, if the
424 user were already presumed to be logged in, and only one simultaneous
425 session were permitted. In these situations, the AAA client might
426 receive both Accept and Reject responses to the same duplicate
427 request, and the outcome will depend on which response arrives first.
428
429 <span class="h3"><a class="selflink" id="section-2.6" href="#section-2.6">2.6</a>. Invalidation of Transport Parameter Estimates</span>
430
431 Congestion control principles [<a href="#ref-Congest" title="&quot;Congestion Avoidance and Control&quot;">Congest</a>],[<a href="./rfc2914" title="&quot;Congestion Control Principles&quot;">RFC2914</a>] require the ability
432 of a transport protocol to respond effectively to congestion, as
433 sensed via increasing delays, packet loss, or explicit congestion
434 notification.
435
436 With network-driven applications, it is possible to respond to
437 congestion on a timescale comparable to the round-trip time (RTT).
438
439 However, with AAA protocols, the time between sends may be longer
440 than the RTT, so that the network conditions can not be assumed to
441
442
443
444 <span class="grey">Aboba &amp; Wood Standards Track [Page 8]</span></pre>
445 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-9" ></span>
446 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
447
448
449 persist between sends. For example, the congestion window may grow
450 during a period in which congestion is being experienced because few
451 packets are sent, limiting the opportunity for feedback. Similarly,
452 after congestion is detected, the congestion window may remain small,
453 even though the network conditions that existed at the time of
454 congestion no longer apply by the time when the next packets are
455 sent. In addition, due to the low sampling interval, estimates of
456 RTT and RTO made via the procedure described in [<a href="./rfc2988" title="&quot;Computing TCP's Retransmission Timer&quot;">RFC2988</a>] may become
457 invalid.
458
459 <span class="h3"><a class="selflink" id="section-2.7" href="#section-2.7">2.7</a>. Inability to Use Fast Re-transmit</span>
460
461 When congestion window validation [<a href="./rfc2861" title="&quot;TCP Congestion Window Validation&quot;">RFC2861</a>] is implemented, the
462 result is that AAA protocols operate much of the time in slow-start
463 with an initial congestion window set to 1 or 2, depending on the
464 implementation [<a href="./rfc3390" title="&quot;Increasing TCP's Initial Window&quot;">RFC3390</a>]. This implies that AAA protocols gain
465 little benefit from the windowing features of reliable transport.
466
467 Since the congestion window is so small, it is generally not possible
468 to receive enough duplicate ACKs (3) to trigger fast re-transmit. In
469 addition, since AAA traffic is two-way, ACKs including data will not
470 count as part of the duplicate ACKs necessary to trigger fast re-
471 transmit. As a result, dropped packets will require a retransmission
472 timeout (RTO).
473
474 <span class="h3"><a class="selflink" id="section-2.8" href="#section-2.8">2.8</a>. Congestion Avoidance</span>
475
476 The law of conservation of packets [<a href="#ref-Congest" title="&quot;Congestion Avoidance and Control&quot;">Congest</a>] suggests that a client
477 should not send another packet into the network until it can be
478 reasonably sure that a packet has exited the network on the same
479 path. In the case of a AAA client, the law suggests that it should
480 not retransmit to the same server or choose another server until it
481 can be reasonably sure that a packet has exited the network on the
482 same path. If the client advances the window as responses arrive,
483 then the client will "self clock", adjusting its transmission rate to
484 the available bandwidth.
485
486 While a AAA client using a reliable transport such as TCP [<a href="./rfc793" title="&quot;Transmission Control Protocol&quot;">RFC793</a>] or
487 SCTP [<a href="./rfc2960" title="&quot;Stream Control Transmission Protocol&quot;">RFC2960</a>] will self-clock when communicating directly with a
488 AAA-server, end-to-end self-clocking is not assured when AAA agents
489 are present.
490
491 As described in the Appendix, AAA agents include Relays, Proxies,
492 Re-directs, Store and Forward proxies, and Transport proxies. Of
493 these agents, only Transport proxies and Re-directs provide a direct
494 transport connection between the AAA client and server, allowing
495 end-to-end self-clocking to occur.
496
497
498
499
500 <span class="grey">Aboba &amp; Wood Standards Track [Page 9]</span></pre>
501 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-10" ></span>
502 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
503
504
505 With Relays, Proxies or Store and Forward proxies, two separate and
506 de-coupled transport connections are used. One connection operates
507 between the AAA client and agent, and another between the agent and
508 server. Since the two transport connections are de-coupled,
509 transport layer ACKs do not flow end-to-end, and self-clocking does
510 not occur.
511
512 For example, consider what happens when the bottleneck exists between
513 a AAA Relay and a AAA server. Self-clocking will occur between the
514 AAA client and AAA Relay, causing the AAA client to adjust its
515 sending rate to the rate at which transport ACKs flow back from the
516 AAA Relay. However, since this rate is higher than the bottleneck
517 bandwidth, the overall system will not self-clock.
518
519 Since there is no direct transport connection between the AAA client
520 and AAA server, the AAA client does not have the ability to estimate
521 end-to-end transport parameters and adjust its sending rate to the
522 bottleneck bandwidth between the Relay and server. As a result, the
523 incoming rate at the AAA Relay can be higher than the rate at which
524 packets can be sent to the AAA server.
525
526 In this case, the end-to-end performance will be determined by
527 details of the agent implementation. In general, the end-to-end
528 transport performance in the presence of Relays, Proxies or Store and
529 Forward proxies will always be worse in terms of delay and packet
530 loss than if the AAA client and server were communicating directly.
531
532 For example, if the agent operates with a large receive buffer, it is
533 possible that a large queue will develop on the receiving side, since
534 the AAA client is able to send packets to the AAA agent more rapidly
535 than the agent can send them to the AAA server. Eventually, the
536 buffer will overflow, causing wholesale packet loss as well as high
537 delay.
538
539 Methods to induce fine-grained coupling between the two transport
540 connections are difficult to implement. One possible solution is for
541 the AAA agent to operate with a receive buffer that is no larger than
542 its send buffer. If this is done, "back pressure" (closing of the
543 receive window) will cause the agent to reduce the AAA client sending
544 rate when the agent send buffer fills. However, unless multiple
545 connections exist between the AAA client and AAA agent, closing of
546 the receive window will affect all traffic sent by the AAA client,
547 even traffic destined to AAA servers where no bottleneck exists.
548 Since multiple connections between a AAA client and agent result in
549 multiplication of the effective slow-start ramp rate, this is not
550 recommended. As a result, use of "back pressure" cannot enable
551 individual AAA client-server conversations to self-clock, and this
552 technique appears impractical for use in AAA.
553
554
555
556 <span class="grey">Aboba &amp; Wood Standards Track [Page 10]</span></pre>
557 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-11" ></span>
558 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
559
560
561 <span class="h3"><a class="selflink" id="section-2.9" href="#section-2.9">2.9</a>. Delayed Acknowledgments</span>
562
563 As described in <a href="#appendix-B">Appendix B</a>, ACKs may comprise as much as half of the
564 traffic generated in a AAA exchange. This occurs because AAA
565 conversations are typically application-driven, and therefore there
566 is frequently not enough traffic to enable ACK piggybacking. As a
567 result, AAA protocols running over TCP or SCTP transport may
568 experience a doubling of traffic as compared with implementations
569 utilizing UDP transport.
570
571 It is typically not possible to address this issue via the sockets
572 API. ACK parameters (such as the value of the delayed ACK timer) are
573 typically fixed by TCP and SCTP implementations and are therefore not
574 tunable by the application.
575
576 <span class="h3"><a class="selflink" id="section-2.10" href="#section-2.10">2.10</a>. Premature Failover</span>
577
578 RADIUS failover implementations are typically based on the concept of
579 primary and secondary servers, in which all traffic flows to the
580 primary server unless it is unavailable. However, the failover
581 algorithm was not specified in [<a href="./rfc2865" title="&quot;Remote Authentication Dial In User Service (RADIUS)&quot;">RFC2865</a>] or [<a href="./rfc2866" title="&quot;RADIUS Accounting&quot;">RFC2866</a>]. As a result,
582 RADIUS failover implementations vary in quality, with some failing
583 over prematurely, violating the law of "conservation of packets".
584
585 Where a Relay, Proxy or Store and Forward proxy is present, the AAA
586 client has no direct connection to a AAA server, and is unable to
587 estimate the end-to-end transport parameters. As a result, a AAA
588 client awaiting an application-layer response from the server has no
589 transport-based mechanism for determining an appropriate failover
590 timer.
591
592 For example, if the path between the AAA agent and server includes a
593 high delay link, or if the AAA server is very heavily loaded, it is
594 possible that the NAS will failover to another agent while packets
595 are still in flight. This violates the principle of "conservation of
596 packets", since the AAA client will inject additional packets into
597 the network before having evidence that a previously sent packet has
598 left the network. Such behavior can result in a worse situation on
599 an already congested link, resulting in congestive collapse
600 [<a href="#ref-Congest" title="&quot;Congestion Avoidance and Control&quot;">Congest</a>].
601
602 <span class="h3"><a class="selflink" id="section-2.11" href="#section-2.11">2.11</a>. Head of Line Blocking</span>
603
604 Head of line blocking occurs during periods of packet loss where the
605 time between sends is shorter than the re-transmission timeout value
606 (RTO). In such situations, packets back up in the send queue until
607
608
609
610
611
612 <span class="grey">Aboba &amp; Wood Standards Track [Page 11]</span></pre>
613 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-12" ></span>
614 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
615
616
617 the lost packet can be successfully re-transmitted. This can be an
618 issue for SCTP when using ordered delivery over a single stream, and
619 for TCP.
620
621 Head of line blocking is typically an issue only on larger NASes.
622 For example, a 48-port NAS with an average inter-packet spacing of 25
623 seconds is unlikely to have an RTO greater than this, unless severe
624 packet loss has been experienced. However, a 2048-port NAS with an
625 average inter-packet spacing of 293 ms may experience head-of-line
626 blocking since the inter-packet spacing is less than the minimum RTO
627 value of 1 second [<a href="./rfc2988" title="&quot;Computing TCP's Retransmission Timer&quot;">RFC2988</a>].
628
629 <span class="h3"><a class="selflink" id="section-2.12" href="#section-2.12">2.12</a>. Connection Load Balancing</span>
630
631 In order to lessen queuing delays and address head of line blocking,
632 a AAA implementation may wish to load balance between connections to
633 multiple destinations. While it is possible to employ dynamic load
634 balancing techniques, this level of sophistication may not be
635 required. In many situations, adequate reliability and load
636 balancing can be achieved via static load balancing, where traffic is
637 distributed between destinations based on static "weights".
638
639 <span class="h2"><a class="selflink" id="section-3" href="#section-3">3</a>. AAA Transport Profile</span>
640
641 In order to address AAA transport issues, it is recommended that AAA
642 protocols make use of standards track as well as experimental
643 techniques. More details are provided in the sections that follow.
644
645 <span class="h3"><a class="selflink" id="section-3.1" href="#section-3.1">3.1</a>. Transport Mappings</span>
646
647 AAA Servers MUST support TCP and SCTP. AAA clients SHOULD support
648 SCTP, but MUST support TCP if SCTP is not available. As support for
649 SCTP improves, it is possible that SCTP support will be required on
650 clients at some point in the future. AAA agents inherit all the
651 obligations of Servers with respect to transport support.
652
653 <span class="h3"><a class="selflink" id="section-3.2" href="#section-3.2">3.2</a>. Use of Nagle Algorithm</span>
654
655 While AAA protocols typically operate in the application-driven
656 regime, there are circumstances in which they are network driven.
657 For example, where an NAS reboots, or where connectivity is restored
658 between an NAS and a AAA agent, it is possible that multiple packets
659 will be available for sending.
660
661 As a result, there are circumstances where the transport-layer
662 batching provided by the Nagle Algorithm (12) is useful, and as a
663 result, AAA implementations running over TCP MUST enable the Nagle
664 algorithm, [<a href="./rfc896" title="&quot;Congestion Control in IP/TCP internetworks&quot;">RFC896</a>]. The Nagle algorithm is not used with SCTP.
665
666
667
668 <span class="grey">Aboba &amp; Wood Standards Track [Page 12]</span></pre>
669 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-13" ></span>
670 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
671
672
673 <span class="h3"><a class="selflink" id="section-3.3" href="#section-3.3">3.3</a>. Multiple Connections</span>
674
675 AAA protocols SHOULD use only a single persistent connection between
676 a AAA client and a AAA agent or server. They SHOULD provide for
677 pipelining of requests, so that more than one request can be in
678 progress at a time. In order to minimize use of inactive connections
679 in roaming situations, a AAA client or agent MAY bring down a
680 connection to a AAA agent or server if the connection has been
681 unutilized (discounting the watchdog) for a certain period of time,
682 which MUST NOT be less than BRINGDOWN_INTERVAL (5 minutes).
683
684 While a AAA client/agent SHOULD only use a single persistent
685 connection to a given AAA agent or server, it MAY have connections to
686 multiple AAA agents or servers. A AAA client/agent connected to
687 multiple agents/servers can treat them as primary/secondary or
688 balance load between them.
689
690 <span class="h3"><a class="selflink" id="section-3.4" href="#section-3.4">3.4</a>. Application Layer Watchdog</span>
691
692 In order to enable AAA implementations to more quickly detect
693 transport and application-layer failures, AAA protocols MUST support
694 an application layer watchdog message.
695
696 The application layer watchdog message enables failover from a peer
697 that has failed, either because it is unreachable or because its
698 applications functions have failed. This is distinct from the
699 purpose of the SCTP heartbeat, which is to enable failover between
700 interfaces. The SCTP heartbeat may enable a failover to another path
701 to reach the same server, but does not address the situation where
702 the server system or the application service has failed. Therefore
703 both mechanisms MAY be used together.
704
705 The watchdog is used in order to enable a AAA client or agent to
706 determine when to resend on another connection. It operates on all
707 open connections and is used to suspend and eventually close
708 connections that are experiencing difficulties. The watchdog is also
709 used to re-open and validate connections that have returned to
710 health. The watchdog may be utilized either within primary/secondary
711 or load balancing configurations. However, it is not intended as a
712 cluster heartbeat mechanism.
713
714 The application layer watchdog is designed to detect failures of the
715 immediate peer, and not to be affected by failures of downstream
716 proxies or servers. This prevents instability in downstream AAA
717 components from propagating upstream. While the receipt of any AAA
718 Response from a peer is taken as evidence that the peer is up, lack
719 of a Response is insufficient to conclude that the peer is down.
720 Since the lack of Response may be the result of problems with a
721
722
723
724 <span class="grey">Aboba &amp; Wood Standards Track [Page 13]</span></pre>
725 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-14" ></span>
726 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
727
728
729 downstream proxy or server, only after failure to respond to the
730 watchdog message can it be determined that the peer is down.
731
732 Since the watchdog algorithm takes any AAA Response into account in
733 determining peer liveness, decreases in the watchdog timer interval
734 do not significantly increase the level of watchdog traffic on
735 heavily loaded networks. This is because watchdog messages do not
736 need to be sent where other AAA Response traffic serves as a constant
737 reminder of peer liveness. Watchdog traffic only increases when AAA
738 traffic is light, and therefore a AAA Response "signal" is not
739 present. Nevertheless, decreasing the timer interval TWINIT does
740 increase the probability of false failover significantly, and so this
741 decision should be made with care.
742
743 <span class="h4"><a class="selflink" id="section-3.4.1" href="#section-3.4.1">3.4.1</a>. Algorithm Overview</span>
744
745 The watchdog behavior is controlled by an algorithm defined in this
746 section. This algorithm is appropriate for use either within
747 primary/secondary or load balancing configurations. Implementations
748 SHOULD implement this algorithm, which operates as follows:
749
750 [<a id="ref-1">1</a>] Watchdog behavior is controlled by a single timer (Tw). The
751 initial value of Tw, prior to jittering is Twinit. The default
752 value of Twinit is 30 seconds. This value was selected because
753 it minimizes the probability that failover will be initiated due
754 to a routing flap, as noted in [<a href="#ref-Paxson" title="&quot;Measurement and Analysis of End-to-End Internet Dynamics&quot;">Paxson</a>].
755
756 While Twinit MAY be set as low as 6 seconds (not including
757 jitter), it MUST NOT be set lower than this. Note that setting
758 such a low value for Twinit is likely to result in an increased
759 probability of duplicates, as well as an increase in spurious
760 failover and failback attempts.
761
762 In order to avoid synchronization behaviors that can occur with
763 fixed timers among distributed systems, each time the watchdog
764 interval is calculated with a jitter by using the Twinit value
765 and randomly adding a value drawn between -2 and 2 seconds.
766 Alternative calculations to create jitter MAY be used. These
767 MUST be pseudo-random, generated by a PRNG seeded as per
768 [<a href="./rfc1750" title="&quot;Randomness Recommendations for Security&quot;">RFC1750</a>].
769
770 [<a id="ref-2">2</a>] When any AAA message is received, Tw is reset. This need not be
771 a response to a watchdog request. Receiving a watchdog response
772 from a peer constitutes activity, and Tw should be reset. If the
773 watchdog timer expires and no watchdog response is pending, then
774 a watchdog message is sent. On sending a watchdog request, Tw is
775 reset.
776
777
778
779
780 <span class="grey">Aboba &amp; Wood Standards Track [Page 14]</span></pre>
781 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-15" ></span>
782 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
783
784
785 Watchdog packets are not retransmitted by the AAA protocol, since
786 AAA protocols run over reliable transports that will handle all
787 retransmissions internally. As a result, a watchdog request is
788 only sent when there is no watchdog response pending.
789
790 [<a id="ref-3">3</a>] If the watchdog timer expires and a watchdog response is pending,
791 then failover is initiated. In order for a AAA client or agent
792 to perform failover procedures, it is necessary to maintain a
793 pending message queue for a given peer. When an answer message
794 is received, the corresponding request is removed from the queue.
795 The Hop-by-Hop Identifier field MAY be used to match the answer
796 with the queued request.
797
798 When failover is initiated, all messages in the queue are sent to
799 an alternate agent, if available. Multiple identical requests or
800 answers may be received as a result of a failover. The
801 combination of an end-to-end identifier and the origin host MUST
802 be used to identify duplicate messages.
803
804 Note that where traffic is heavy, the application layer watchdog
805 can take as long as 2Tw to determine that a peer has gone down.
806 For peers receiving a high volume of AAA Requests, AAA Responses
807 will continually reset the timer, so that after a failure it will
808 take Tw for the lack of traffic to be noticed, and for the
809 watchdog message to be sent. Another Tw will elapse before
810 failover is initiated.
811
812 On a lightly loaded network without much AAA Response traffic,
813 the watchdog timer will typically expire without being reset, so
814 that a watchdog response will be outstanding and failover will be
815 initiated after only a single timer interval has expired.
816
817 [<a id="ref-4">4</a>] The client MUST NOT close the primary connection until the
818 primary's watchdog timer has expired at least twice without a
819 response (note that the watchdog is not sent a second time,
820 however). Once this has occurred, the client SHOULD cause a
821 transport reset or close to be done on the connection.
822
823 Once the primary connection has failed, subsequent requests are
824 sent to the alternate server until the watchdog timer on the
825 primary connection is reset.
826
827 Suspension of the primary connection prevents flapping between
828 primary and alternate connections, and ensures that failover
829 behavior remains consistent. The application may not receive a
830 response to the watchdog request message due to a connectivity
831 problem, in which case a transport layer ACK will not have been
832 received, or the lack of response may be due to an application
833
834
835
836 <span class="grey">Aboba &amp; Wood Standards Track [Page 15]</span></pre>
837 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-16" ></span>
838 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
839
840
841 problem. Without transport layer visibility, the application is
842 unable to tell the difference, and must behave conservatively.
843
844 In situations where no transport layer ACK is received on the
845 primary connection after multiple re-transmissions, the RTO will
846 be exponentially backed off as described in [<a href="./rfc2988" title="&quot;Computing TCP's Retransmission Timer&quot;">RFC2988</a>]. Due to
847 Karn's algorithm as implemented in SCTP and TCP, the RTO
848 estimator will not be reset until another ACK is received in
849 response to a non-re-transmitted request. Thus, in cases where
850 the problem occurs at the transport layer, after the client fails
851 over to the alternate server, the RTO of the primary will remain
852 at a high value unless an ACK is received on the primary
853 connection.
854
855 In the case where the problem occurs at the transport layer,
856 subsequent requests sent on the primary connection will not
857 receive the same service as was originally provided. For
858 example, instead of failover occurring after 3 retransmissions,
859 failover might occur without even a single retransmission if RTO
860 has been sufficiently backed off. Of course, if the lack of a
861 watchdog response was due to an application layer problem, then
862 RTO will not have been backed off. However, without transport
863 layer visibility, there is no way for the application to know
864 this.
865
866 Suspending use of the primary connection until a response to a
867 watchdog message is received guarantees that the RTO timer will
868 have been reset before the primary connection is reused. If no
869 response is received after the second watchdog timer expiration,
870 then the primary connection is closed and the suspension becomes
871 permanent.
872
873 [<a id="ref-5">5</a>] While the connection is in the closed state, the AAA client MUST
874 NOT attempt to send further watchdog messages on the connection.
875 However, after the connection is closed, the AAA client continues
876 to periodically attempt to reopen the connection.
877
878 The AAA client SHOULD wait for the transport layer to report
879 connection failure before attempting again, but MAY choose to
880 bound this wait time by the watchdog interval, Tw. If the
881 connection is successfully opened, then the watchdog message is
882 sent. Once three watchdog messages have been sent and responded
883 to, the connection is returned to service, and transactions are
884 once again sent over it. Connection validation via receipt of
885 multiple watchdogs is not required when a connection is initially
886 brought up -- in this case, the connection can immediately be put
887 into service.
888
889
890
891
892 <span class="grey">Aboba &amp; Wood Standards Track [Page 16]</span></pre>
893 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-17" ></span>
894 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
895
896
897 [<a id="ref-6">6</a>] When using SCTP as a transport, it is not necessary to disable
898 SCTP's transport-layer heartbeats. However, if AAA
899 implementations have access to SCTP's heartbeat parameters, they
900 MAY chose to ensure that SCTP's heartbeat interval is longer than
901 the AAA watchdog interval, Tw. This will ensure that alternate
902 paths are still probed by SCTP, while the primary path has a
903 minimum of heartbeat redundancy.
904
905 <span class="h4"><a class="selflink" id="section-3.4.2" href="#section-3.4.2">3.4.2</a>. Primary/Secondary Failover Support</span>
906
907 The watchdog timer MAY be integrated with primary/secondary style
908 failover so as to provide improved reliability and basic load
909 balancing. In order to balance load among multiple AAA servers, each
910 AAA server is designated the primary for a portion of the clients,
911 and designated as secondaries of varying priority for the remainder.
912 In this way, load can be balanced among the AAA servers.
913
914 Within primary/secondary configurations, the watchdog timer operates
915 as follows:
916
917 [<a id="ref-1">1</a>] Assume that each client or agent is initially configured with a
918 single primary agent or server, and one or more secondary
919 connections.
920
921 [<a id="ref-2">2</a>] The watchdog mechanism is used to suspend and eventually close
922 primary connections that are experiencing difficulties. It is
923 also used to re-open and validate connections that have returned
924 to health.
925
926 [<a id="ref-3">3</a>] Once a secondary is promoted to primary status, either on a
927 temporary or permanent basis, the next server on the list of
928 secondaries is promoted to fill the open secondary slot.
929
930 [<a id="ref-4">4</a>] The client or agent periodically attempts to re-open closed
931 connections, so that it is possible that a previously closed
932 connection can be returned to service and become eligible for use
933 again. Implementations will typically retain a limit on the
934 number of connections open at a time, so that once a previously
935 closed connection is brought online again, the lowest priority
936 secondary connection will be closed. In order to prevent
937 periodic closing and re-opening of secondary connections, it is
938 recommended that functioning connections remain open for a
939 minimum of 5 minutes.
940
941 [<a id="ref-5">5</a>] In order to enable diagnosis of failover behavior, it is
942 recommended that a table of failover events be kept within the
943 MIB. These failover events SHOULD include appropriate
944 transaction identifiers so that client and server data can be
945
946
947
948 <span class="grey">Aboba &amp; Wood Standards Track [Page 17]</span></pre>
949 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-18" ></span>
950 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
951
952
953 compared, providing insight into the cause of the problem
954 (transport or application layer).
955
956 <span class="h4"><a class="selflink" id="section-3.4.3" href="#section-3.4.3">3.4.3</a>. Connection Load Balancing</span>
957
958 Primary/secondary failover is capable of providing improved
959 resilience and basic load balancing. However, it does not address
960 TCP head of line blocking, since only a single connection is in use
961 at a time.
962
963 A AAA client or agent maintaining connections to multiple agents or
964 servers MAY load balance between them. Establishing connections to
965 multiple agents or servers reduces, but does not eliminate, head of
966 line blocking issues experienced on TCP connections. This issue does
967 not exist with SCTP connections utilizing multiple streams.
968
969 In connection load balancing configurations, the application watchdog
970 operates as follows:
971
972 [<a id="ref-1">1</a>] Assume that each client or agent is initially configured with
973 connections to multiple AAA agents or servers, with one
974 connection between a given client/agent and an agent/server.
975
976 [<a id="ref-2">2</a>] In static load balancing, transactions are apportioned among the
977 connections based on the total number of connections and a
978 "weight" assigned to each connection. Pearson's hash [<a href="./rfc3074" title="&quot;DHC Load Balancing Algorithm&quot;">RFC3074</a>]
979 applied to the NAI [<a href="./rfc2486" title="&quot;The Network Access Identifier&quot;">RFC2486</a>] can be used to determine which
980 connection will handle a given transaction. Hashing on the NAI
981 provides highly granular load balancing, while ensuring that all
982 traffic for a given conversation will be sent to the same agent
983 or server. In dynamic load balancing, the value of the "weight"
984 can vary based on conditions such as AAA server load. Such
985 techniques, while sophisticated, are beyond the scope of this
986 document.
987
988 [<a id="ref-3">3</a>] Transactions are distributed to connections based on the total
989 number of available connections and their weights. A change in
990 the number of available connections forces recomputation of the
991 hash table. In order not to cause conversations in progress to
992 be switched to new destinations, on recomputation, a transitional
993 period is required in which both old and new hash tables are
994 needed in order to permit aging out of conversations in progress.
995 Note that this requires a way to easily determine whether a
996 Request represents a new conversation or the continuation of an
997 existing conversation. As a result, removing and adding of
998 connections is an expensive operation, and it is recommended that
999 the hash table only be recomputed once a connection is closed or
1000 returned to service.
1001
1002
1003
1004 <span class="grey">Aboba &amp; Wood Standards Track [Page 18]</span></pre>
1005 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-19" ></span>
1006 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1007
1008
1009 Suspended connections, although they are not used, do not force
1010 hash table reconfiguration until they are closed. Similarly,
1011 re-opened connections not accumulating sufficient watchdog
1012 responses do not force a reconfiguration until they are returned
1013 to service.
1014
1015 While a connection is suspended, transactions that were to have
1016 been assigned to it are instead assigned to the next available
1017 server. While this results in a momentary imbalance, it is felt
1018 that this is a relatively small price to pay in order to reduce
1019 hash table thrashing.
1020
1021 [<a id="ref-4">4</a>] In order to enable diagnosis of load balancing behavior, it is
1022 recommended that in addition to a table of failover events, a
1023 table of statistics be kept on each client, indexed by a AAA
1024 server. That way, the effectiveness of the load balancing
1025 algorithm can be evaluated.
1026
1027 <span class="h3"><a class="selflink" id="section-3.5" href="#section-3.5">3.5</a>. Duplicate Detection</span>
1028
1029 Multiple facilities are required to enable duplicate detection.
1030 These include session identifiers as well as hop-by-hop and end-to-
1031 end message identifiers. Hop-by-hop identifiers whose value may
1032 change at each hop are not sufficient, since a AAA server may receive
1033 the same message from multiple agents. For example, a AAA client can
1034 send a request to Agent1, then failover and resend the request to
1035 Agent2; both agents forward the request to the home AAA server, with
1036 different hop-by-hop identifiers. A Session Identifier is
1037 insufficient as it does not distinguish different messages for the
1038 the same session.
1039
1040 Proper treatment of the end-to-end message identifier ensures that
1041 AAA operations are idempotent. For example, without an end-to-end
1042 identifier, a AAA server keeping track of simultaneous logins might
1043 send an Accept in response to an initial Request, and then a Reject
1044 in response to a duplicate Request (where the user was allowed only
1045 one simultaneous login). Depending on which Response arrived first,
1046 the user might be allowed access or not.
1047
1048 However, if the server were to store the end-to-end message
1049 identifier along with the simultaneous login information, then the
1050 duplicate Request (which utilizes the same end-to-end message
1051 identifier) could be identified and the correct response could be
1052 returned.
1053
1054
1055
1056
1057
1058
1059
1060 <span class="grey">Aboba &amp; Wood Standards Track [Page 19]</span></pre>
1061 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-20" ></span>
1062 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1063
1064
1065 <span class="h3"><a class="selflink" id="section-3.6" href="#section-3.6">3.6</a>. Invalidation of Transport Parameter Estimates</span>
1066
1067 In order to address invalidation of transport parameter estimates,
1068 AAA protocol implementations MAY utilize Congestion Window Validation
1069 [<a href="./rfc2861" title="&quot;TCP Congestion Window Validation&quot;">RFC2861</a>] and RTO validation when using TCP. This specification also
1070 recommends a procedure for RTO validation.
1071
1072 [<a id="ref-RFC2581">RFC2581</a>] and [<a href="./rfc2861" title="&quot;TCP Congestion Window Validation&quot;">RFC2861</a>] both recommend that a connection go into
1073 slow-start after a period where no traffic has been sent within the
1074 RTO interval. [<a href="./rfc2861" title="&quot;TCP Congestion Window Validation&quot;">RFC2861</a>] recommends only increasing the congestion
1075 window if it was full when the ACK arrived. The congestion window is
1076 reduced by half once every RTO interval if no traffic is received.
1077
1078 When Congestion Window Validation is used, the congestion window will
1079 not build during application-driven periods, and instead will be
1080 decayed. As a result, AAA applications operating within the
1081 application-driven regime will typically run with a congestion window
1082 equal to the initial window much of the time, operating in "perpetual
1083 slowstart".
1084
1085 During periods in which AAA behavior is application-driven this will
1086 have no effect. Since the time between packets will be larger than
1087 RTT, AAA will operate with an effective congestion window equal to
1088 the initial window. However, during network-driven periods, the
1089 effect will be to space out sending of AAA packets. Thus instead of
1090 being able to send a large burst of packets into the network, a
1091 client will need to wait several RTTs as the congestion window builds
1092 during slow-start.
1093
1094 For example, a client operating over TCP with an initial window of 2,
1095 with 35 AAA requests to send would take approximately 6 RTTs to send
1096 them, as the congestion window builds during slow start: 2, 3, 3, 6,
1097 9, 12. After the backlog is cleared, the implementation will once
1098 again be application-driven and the congestion window size will
1099 decay. If the client were using SCTP, the number of RTTs needed to
1100 transmit all requests would usually be less, and would depend on the
1101 size of the requests, since SCTP tracks the progress for the opening
1102 of the congestion window by bytes, not segments.
1103
1104 Note that [<a href="./rfc2861" title="&quot;TCP Congestion Window Validation&quot;">RFC2861</a>] and [<a href="./rfc2988" title="&quot;Computing TCP's Retransmission Timer&quot;">RFC2988</a>] do not address the issue of RTO
1105 validation. This is also a problem, particularly when the Congestion
1106 Manager [<a href="./rfc3124" title="&quot;The Congestion Manager&quot;">RFC3124</a>] is implemented. During periods of high packet
1107 loss, the RTO may be repeatedly increased via exponential back-off,
1108 and may attain a high value. Due to lack of timely feedback on RTT
1109 and RTO during application-driven periods, the high RTO estimate may
1110 persist long after the conditions that generated it have dissipated.
1111
1112
1113
1114
1115
1116 <span class="grey">Aboba &amp; Wood Standards Track [Page 20]</span></pre>
1117 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-21" ></span>
1118 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1119
1120
1121 RTO validation MAY be used to address this issue for TCP, via the
1122 following procedure:
1123
1124 After the congestion window is decayed according to [<a href="./rfc2861" title="&quot;TCP Congestion Window Validation&quot;">RFC2861</a>],
1125 reset the estimated RTO to 3 seconds. After the next packet comes
1126 in, re-calculate RTTavg, RTTdev, and RTO according to the method
1127 described in [<a href="./rfc2581" title="&quot;TCP Congestion Control&quot;">RFC2581</a>].
1128
1129 To address this issue for SCTP, AAA implementations SHOULD use SCTP
1130 heartbeats. [<a href="./rfc2960" title="&quot;Stream Control Transmission Protocol&quot;">RFC2960</a>] states that heartbeats should be enabled by
1131 default, with an interval of 30 seconds. If this interval proves to
1132 be too long to resolve this issue, AAA implementations MAY reduce the
1133 heartbeat interval.
1134
1135 <span class="h3"><a class="selflink" id="section-3.7" href="#section-3.7">3.7</a>. Inability to Use Fast Re-Transmit</span>
1136
1137 When Congestion Window Validation [<a href="./rfc2861" title="&quot;TCP Congestion Window Validation&quot;">RFC2861</a>] is used, AAA
1138 implementations will operate with a congestion window equal to the
1139 initial window much of the time. As a result, the window size will
1140 often not be large enough to enable use of fast re-transmit for TCP.
1141 In addition, since AAA traffic is two-way, ACKs carrying data will
1142 not count towards triggering fast re-transmit. SCTP is less likely
1143 to encounter this issue, so the measures described below apply to
1144 TCP.
1145
1146 To address this issue, AAA implementations SHOULD support selective
1147 acknowledgement as described in [<a href="./rfc2018" title="&quot;TCP Selective Acknowledgment Options&quot;">RFC2018</a>] and [<a href="./rfc2883" title="&quot;An Extension to the Selective Acknowledgment (SACK) Option for TCP&quot;">RFC2883</a>]. AAA
1148 implementations SHOULD also implement Limited Transmit for TCP, as
1149 described in [<a href="./rfc3042" title="&quot;Enhancing TCP's Loss Recovery Using Limited Transmit&quot;">RFC3042</a>]. Rather than reducing the number of duplicate
1150 ACKs required for triggering fast recovery, which would increase the
1151 number of inappropriate re-transmissions, Limited Transmit enables
1152 the window size be increased, thus enabling the sending of additional
1153 packets which in turn may trigger fast re-transmit without a change
1154 to the algorithm.
1155
1156 However, if congestion window validation [<a href="./rfc2861" title="&quot;TCP Congestion Window Validation&quot;">RFC2861</a>] is implemented,
1157 this proposal will only have an effect in situations where the time
1158 between packets is less than the estimated retransmission timeout
1159 (RTO). If the time between packets is greater than RTO, additional
1160 packets will typically not be available for sending so as to take
1161 advantage of the increased window size. As a result, AAA protocols
1162 will typically operate with the lowest possible congestion window
1163 size, resulting in a re-transmission timeout for every lost packet.
1164
1165
1166
1167
1168
1169
1170
1171
1172 <span class="grey">Aboba &amp; Wood Standards Track [Page 21]</span></pre>
1173 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-22" ></span>
1174 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1175
1176
1177 <span class="h3"><a class="selflink" id="section-3.8" href="#section-3.8">3.8</a>. Head of Line Blocking</span>
1178
1179 TCP inherently does not provide a solution to the head-of-line
1180 blocking problem, although its effects can be lessened by
1181 implementation of Limited Transmit [<a href="./rfc3042" title="&quot;Enhancing TCP's Loss Recovery Using Limited Transmit&quot;">RFC3042</a>], and connection load
1182 balancing.
1183
1184 <span class="h4"><a class="selflink" id="section-3.8.1" href="#section-3.8.1">3.8.1</a>. Using SCTP Streams to Prevent Head of Line Blocking</span>
1185
1186 Each AAA node SHOULD distribute its messages evenly across the range
1187 of SCTP streams that it and its peer have agreed upon. (A lost
1188 message in one stream will not cause any other streams to block.) A
1189 trivial and effective implementation of this simply increments a
1190 counter for the stream ID to send on. When the counter reaches the
1191 maximum number of streams for the association, it resets to 0.
1192
1193 AAA peers MUST be able to accept messages on any stream. Note that
1194 streams are used *solely* to prevent head-of-the-line blocking. All
1195 identifying information is carried within the Diameter payload.
1196 Messages distributed across multiple streams may not be received in
1197 the order they are sent.
1198
1199 SCTP peers can allocate up to 65535 streams for an association. The
1200 cost for idle streams may or may not be zero, depending on the
1201 implementation, and the cost for non-idle streams is always greater
1202 than 0. So administrators may wish to limit the number of possible
1203 streams on their diameter nodes according to the resources (i.e.
1204 memory, CPU power, etc.) of a particular node.
1205
1206 On a Diameter client, the number of streams may be determined by the
1207 maximum number of peak users on the NAS. If a stream is available
1208 per user, then this should be sufficient to prevent head-of-line
1209 blocking. On a Diameter proxy, the number of streams may be
1210 determined by the maximum number of peak sessions in progress from
1211 that proxy to each downstream AAA server.
1212
1213 Stream IDs do not need to be preserved by relay agents. This
1214 simplifies implementation, as agents can easily handle forwarding
1215 between two associations with different numbers of streams. For
1216 example, consider the following case, where a relay server DRL
1217 forwards messages between a NAS and a home server, HMS. The NAS and
1218 DRL have agreed upon 1000 streams for their association, and DRL and
1219 HMS have agreed upon 2000 streams for their association. The
1220 following figure shows the message flow from NAS to HMS via DRL, and
1221 the stream ID assignments for each message:
1222
1223
1224
1225
1226
1227
1228 <span class="grey">Aboba &amp; Wood Standards Track [Page 22]</span></pre>
1229 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-23" ></span>
1230 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1231
1232
1233 +------+ +------+ +------+
1234 | | | | | |
1235 | NAS | ---------&gt; | DRL | ---------&gt; | HMS |
1236 | | | | | |
1237 +------+ 1000 streams +------+ 2000 streams +------+
1238
1239 msg 1: str id 0 msg 1: str id 0
1240 msg 2: str id 1 msg 2: str id 1
1241 ...
1242 msg 1000: str id 999 msg 1000: str id 999
1243 msg 1001: str id 0 msg 1001: str id 1000
1244
1245 DRL can forward messages 1 through 1000 to HMS using the same stream
1246 ID that NAS used to send to DRL. However, since the NAS / DRL
1247 association has only 1000 streams, NAS wraps around to stream ID 0
1248 when sending message 1001. The DRL / HMS association, on the other
1249 hand, has 2000 streams, so DRL can reassign message 1001 to stream ID
1250 1000 when forwarding it on to HMS.
1251
1252 This distribution scheme acts like a hash table. It is possible, yet
1253 unlikely, that two messages will end up in the same stream, and even
1254 less likely that there will be message loss resulting in blocking
1255 when this happens. If it does turn out to be a problem, local
1256 administrators can increase the number of streams on their nodes to
1257 improve performance.
1258
1259 <span class="h3"><a class="selflink" id="section-3.9" href="#section-3.9">3.9</a>. Congestion Avoidance</span>
1260
1261 In order to improve upon default timer estimates, AAA implementations
1262 MAY implement the Congestion Manager (CM) [<a href="./rfc3124" title="&quot;The Congestion Manager&quot;">RFC3124</a>]. CM is an end-
1263 system module that:
1264
1265 (i) Enables an ensemble of multiple concurrent streams from a
1266 sender destined to the same receiver and sharing the same
1267 congestion properties to perform proper congestion avoidance
1268 and control, and
1269
1270 (ii) Allows applications to easily adapt to network congestion.
1271
1272 The CM helps integrate congestion management across all applications
1273 and transport protocols. The CM maintains congestion parameters
1274 (available aggregate and per-stream bandwidth, per-receiver round-
1275 trip times, etc.) and exports an API that enables applications to
1276 learn about network characteristics, pass information to the CM,
1277 share congestion information with each other, and schedule data
1278 transmissions.
1279
1280
1281
1282
1283
1284 <span class="grey">Aboba &amp; Wood Standards Track [Page 23]</span></pre>
1285 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-24" ></span>
1286 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1287
1288
1289 The CM enables the AAA application to access transport parameters
1290 (RTTavg, RTTdev) via callbacks. RTO estimates are currently not
1291 available via the callback interface, though they probably should be.
1292 Where available, transport parameters SHOULD be used to improve upon
1293 default timer values.
1294
1295 <span class="h3"><a class="selflink" id="section-3.10" href="#section-3.10">3.10</a>. Premature Failover</span>
1296
1297 Premature failover is prevented by the watchdog functionality
1298 described above. If the next hop does not return a reply, the AAA
1299 client will send a watchdog message to it to verify liveness. If a
1300 watchdog reply is received, then the AAA client will know that the
1301 next hop server is functioning at the application layer. As a
1302 result, it is only necessary to provide terminal error messages, such
1303 as the following:
1304
1305 "Busy": agent/Server too busy to handle additional requests, NAS
1306 should failover all requests to another agent/server.
1307
1308 "Can't Locate": agent can't locate the AAA server for the
1309 indicated realm; NAS should failover that request to another
1310 proxy.
1311
1312 "Can't Forward": agent has tried both primary and secondary AAA
1313 servers with no response; NAS should failover the request to
1314 another agent.
1315
1316 Note that these messages differ in their scope. The "Busy" message
1317 tells the NAS that the agent/server is too busy for ANY request. The
1318 "Can't Locate" and "Can't Forward" messages indicate that the
1319 ultimate destination cannot be reached or isn't responding, implying
1320 per-request failover.
1321
1322 <span class="h2"><a class="selflink" id="section-4" href="#section-4">4</a>. Security Considerations</span>
1323
1324 Since AAA clients, agents and servers serve as network access
1325 gatekeepers, they are tempting targets for attackers. General
1326 security considerations concerning TCP congestion control are
1327 discussed in [<a href="./rfc2581" title="&quot;TCP Congestion Control&quot;">RFC2581</a>]. However, there are some additional
1328 considerations that apply to this specification.
1329
1330 By enabling failover between AAA agents, this specification improves
1331 the resilience of AAA applications. However, it may also open
1332 avenues for denial of service attacks.
1333
1334 The failover algorithm is driven by lack of response to AAA requests
1335 and watchdog packets. On a lightly loaded network where AAA
1336 responses would not be received prior to expiration of the watchdog
1337
1338
1339
1340 <span class="grey">Aboba &amp; Wood Standards Track [Page 24]</span></pre>
1341 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-25" ></span>
1342 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1343
1344
1345 timer, an attacker can swamp the network, causing watchdog packets to
1346 be dropped. This will cause the AAA client to switch to another AAA
1347 agent, where the attack can be repeated. By causing the AAA client
1348 to cycle between AAA agents, service can be denied to users desiring
1349 network access.
1350
1351 Where TLS [<a href="./rfc2246" title="&quot;The TLS Protocol Version 1.0&quot;">RFC2246</a>] is being used to provide AAA security, there will
1352 be a vulnerability to spoofed reset packets, as well as other
1353 transport layer denial of service attacks (e.g. SYN flooding). Since
1354 SCTP offers improved denial of service resilience compared with TCP,
1355 where AAA applications run over SCTP, this can be mitigated to some
1356 extent.
1357
1358 Where IPsec [<a href="./rfc2401" title="&quot;Security Architecture for the Internet Protocol&quot;">RFC2401</a>] is used to provide security, it is important
1359 that IPsec policy require IPsec on incoming packets. In order to
1360 enable a AAA client to determine what security mechanisms are in use
1361 on an agent or server without prior knowledge, it may be tempting to
1362 initiate a connection in the clear, and then to have the AAA agent
1363 respond with IKE [<a href="./rfc2409" title="&quot;The Internet Key Exchange (IKE)&quot;">RFC2409</a>]. While this approach minimizes required
1364 client configuration, it increases the vulnerability to denial of
1365 service attack, since a connection request can now not only tie up
1366 transport resources, but also resources within the IKE
1367 implementation.
1368
1369 <span class="h2"><a class="selflink" id="section-5" href="#section-5">5</a>. IANA Considerations</span>
1370
1371 This document does not create any new number spaces for IANA
1372 administration.
1373
1374 References
1375
1376 <span class="h3"><a class="selflink" id="section-6.1" href="#section-6.1">6.1</a>. Normative References</span>
1377
1378 [<a id="ref-RFC793">RFC793</a>] Postel, J., "Transmission Control Protocol", STD 7, <a href="./rfc793">RFC</a>
1379 <a href="./rfc793">793</a>, September 1981.
1380
1381 [<a id="ref-RFC896">RFC896</a>] Nagle, J., "Congestion Control in IP/TCP internetworks",
1382 <a href="./rfc896">RFC 896</a>, January 1984.
1383
1384 [<a id="ref-RFC1750">RFC1750</a>] Eastlake, D., Crocker, S. and J. Schiller, "Randomness
1385 Recommendations for Security", <a href="./rfc1750">RFC 1750</a>, December 1994.
1386
1387 [<a id="ref-RFC2018">RFC2018</a>] Mathis, M., Mahdavi, J., Floyd, S. and A. Romanow, "TCP
1388 Selective Acknowledgment Options", <a href="./rfc2018">RFC 2018</a>, October 1996.
1389
1390 [<a id="ref-RFC2119">RFC2119</a>] Bradner, S., "Key words for use in RFCs to Indicate
1391 Requirement Levels", <a href="https://www.rfc-editor.org/bcp/bcp14">BCP 14</a>, <a href="./rfc2119">RFC 2119</a>, March 1997.
1392
1393
1394
1395
1396 <span class="grey">Aboba &amp; Wood Standards Track [Page 25]</span></pre>
1397 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-26" ></span>
1398 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1399
1400
1401 [<a id="ref-RFC2486">RFC2486</a>] Aboba, B. and M. Beadles, "The Network Access Identifier",
1402 <a href="./rfc2486">RFC 2486</a>, January 1999.
1403
1404 [<a id="ref-RFC2581">RFC2581</a>] Allman, M., Paxson, V. and W. Stevens, "TCP Congestion
1405 Control", <a href="./rfc2581">RFC 2581</a>, April 1999.
1406
1407 [<a id="ref-RFC2883">RFC2883</a>] Floyd, S., Mahdavi, J., Mathis, M., Podolsky, M. and A.
1408 Romanow, "An Extension to the Selective Acknowledgment
1409 (SACK) Option for TCP", <a href="./rfc2883">RFC 2883</a>, July 2000.
1410
1411 [<a id="ref-RFC2960">RFC2960</a>] Stewart, R., Xie, Q., Morneault, K., Sharp, C.,
1412 Schwarzbauer, H., Taylor, T., Rytina, I., Kalla, M., Zhang,
1413 L. and V. Paxson, "Stream Control Transmission Protocol",
1414 <a href="./rfc2960">RFC 2960</a>, October 2000.
1415
1416 [<a id="ref-RFC2988">RFC2988</a>] Paxson, V. and M. Allman, "Computing TCP's Retransmission
1417 Timer", <a href="./rfc2988">RFC 2988</a>, November 2000.
1418
1419 [<a id="ref-RFC3042">RFC3042</a>] Allman, M., Balakrishnan H. and S. Floyd, "Enhancing TCP's
1420 Loss Recovery Using Limited Transmit", <a href="./rfc3042">RFC 3042</a>, January
1421 2001.
1422
1423 [<a id="ref-RFC3074">RFC3074</a>] Volz, B., Gonczi, S., Lemon, T. and R. Stevens, "DHC Load
1424 Balancing Algorithm", <a href="./rfc3074">RFC 3074</a>, February 2001.
1425
1426 [<a id="ref-RFC3124">RFC3124</a>] Balakrishnan, H. and S. Seshan, "The Congestion Manager",
1427 <a href="./rfc3124">RFC 3124</a>, June 2001.
1428
1429 <span class="h3"><a class="selflink" id="section-6.2" href="#section-6.2">6.2</a>. Informative References</span>
1430
1431 [<a id="ref-RFC2246">RFC2246</a>] Dierks, T. and C. Allen, "The TLS Protocol Version 1.0",
1432 <a href="./rfc2246">RFC 2246</a>, January 1999.
1433
1434 [<a id="ref-RFC2401">RFC2401</a>] Atkinson, R. and S. Kent, "Security Architecture for the
1435 Internet Protocol", <a href="./rfc2401">RFC 2401</a>, November 1998.
1436
1437 [<a id="ref-RFC2409">RFC2409</a>] Harkins, D. and D. Carrel, "The Internet Key Exchange
1438 (IKE)", <a href="./rfc2409">RFC 2409</a>, November 1998.
1439
1440 [<a id="ref-RFC2607">RFC2607</a>] Aboba, B. and J. Vollbrecht, "Proxy Chaining and Policy
1441 Implementation in Roaming", <a href="./rfc2607">RFC 2607</a>, June 1999.
1442
1443 [<a id="ref-RFC2861">RFC2861</a>] Handley, M., Padhye, J. and S. Floyd, "TCP Congestion
1444 Window Validation", <a href="./rfc2861">RFC 2861</a>, June 2000.
1445
1446 [<a id="ref-RFC2865">RFC2865</a>] Rigney, C., Willens, S., Rubens, A. and W. Simpson, "Remote
1447 Authentication Dial In User Service (RADIUS)", <a href="./rfc2865">RFC 2865</a>,
1448 June 2000.
1449
1450
1451
1452 <span class="grey">Aboba &amp; Wood Standards Track [Page 26]</span></pre>
1453 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-27" ></span>
1454 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1455
1456
1457 [<a id="ref-RFC2866">RFC2866</a>] Rigney, C., "RADIUS Accounting", <a href="./rfc2866">RFC 2866</a>, June 2000.
1458
1459 [<a id="ref-RFC2914">RFC2914</a>] Floyd, S., "Congestion Control Principles", <a href="https://www.rfc-editor.org/bcp/bcp41">BCP 41</a>, <a href="./rfc2914">RFC</a>
1460 <a href="./rfc2914">2914</a>, September 2000.
1461
1462 [<a id="ref-RFC2975">RFC2975</a>] Aboba, B., Arkko, J. and D. Harrington, "Introduction to
1463 Accounting Management", <a href="./rfc2975">RFC 2975</a>, June 2000.
1464
1465 [<a id="ref-RFC3390">RFC3390</a>] Allman, M., Floyd, S. and C. Partridge, "Increasing TCP's
1466 Initial Window", <a href="./rfc3390">RFC 3390</a>, October 2002.
1467
1468 [<a id="ref-Congest">Congest</a>] Jacobson, V., "Congestion Avoidance and Control", Computer
1469 Communication Review, vol. 18, no. 4, pp. 314-329, Aug.
1470 1988. <a href="ftp://ftp.ee.lbl.gov/papers/congavoid.ps.Z">ftp://ftp.ee.lbl.gov/papers/congavoid.ps.Z</a>
1471
1472 [<a id="ref-Paxson">Paxson</a>] Paxson, V., "Measurement and Analysis of End-to-End
1473 Internet Dynamics", Ph.D. Thesis, Computer Science
1474 Division, University of California, Berkeley, April 1997.
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508 <span class="grey">Aboba &amp; Wood Standards Track [Page 27]</span></pre>
1509 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-28" ></span>
1510 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1511
1512
1513 Appendix A - Detailed Watchdog Algorithm
1514
1515 In this Appendix, the memory control structure that contains all
1516 information regarding a specific peer is referred to as a Peer
1517 Control Block, or PCB. The PCB contains the following fields:
1518
1519 Status:
1520 OKAY: The connection is up
1521 SUSPECT: Failover has been initiated on the connection.
1522 DOWN: Connection has been closed.
1523 REOPEN: Attempting to reopen a closed connection
1524 INITIAL: The initial state of the pcb when it is first created.
1525 The pcb has never been opened.
1526
1527 Variables:
1528 Pending: Set to TRUE if there is an outstanding unanswered
1529 watchdog request
1530 Tw: Watchdog timer value
1531 NumDWA: Number of DWAs received during REOPEN
1532
1533 Tw is the watchdog timer, measured in seconds. Every second, Tw is
1534 decremented. When it reaches 0, the OnTimerElapsed event (see below)
1535 is invoked. Pseudo-code for the algorithm is included on the
1536 following pages.
1537
1538 SetWatchdog()
1539 {
1540 /*
1541 SetWatchdog() is called whenever it is necessary
1542 to reset the watchdog timer Tw. The value of the
1543 watchdog timer is calculated based on the default
1544 initial value TWINIT and a jitter ranging from
1545 -2 to 2 seconds. The default for TWINIT is 30 seconds,
1546 and MUST NOT be set lower than 6 seconds.
1547 */
1548 Tw=TWINIT -2.0 + 4.0 * random() ;
1549 SetTimer(Tw) ;
1550 return ;
1551 }
1552
1553 /*
1554 OnReceive() is called whenever a message
1555 is received from the peer. This message MAY
1556 be a request or an answer, and can include
1557 DWR and DWA messages. Pending is assumed to
1558 be a global variable.
1559 */
1560 OnReceive(pcb, msgType)
1561
1562
1563
1564 <span class="grey">Aboba &amp; Wood Standards Track [Page 28]</span></pre>
1565 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-29" ></span>
1566 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1567
1568
1569 {
1570 if (msgType == DWA) {
1571 Pending = FALSE;
1572 }
1573 switch (pcb-&gt;Status){
1574 case OKAY:
1575 SetWatchdog();
1576 break;
1577 case SUSPECT:
1578 pcb-&gt;Status = OKAY;
1579 Failback(pcb);
1580 SetWatchdog();
1581 break;
1582 case REOPEN:
1583 if (msgType == DWA) {
1584 NumDWA++;
1585 if (NumDWA == 3) {
1586 pcb-&gt;status = OKAY;
1587 Failback();
1588 }
1589 } else {
1590 Throwaway(received packet);
1591 }
1592 break;
1593 case INITIAL:
1594 case DOWN:
1595 Throwaway(received packet);
1596 break;
1597 default:
1598 Error("Shouldn't be here!");
1599 break;
1600 }
1601 }
1602
1603 /*
1604 OnTimerElapsed() is called whenever Tw reaches zero (0).
1605 */
1606 OnTimerElapsed(pcb)
1607 {
1608 switch (pcb-&gt;status){
1609 case OKAY:
1610 if (!Pending) {
1611 SendWatchdog(pcb);
1612 SetWatchdog();
1613 Pending = TRUE;
1614 break;
1615 }
1616 pcb-&gt;status = SUSPECT;
1617
1618
1619
1620 <span class="grey">Aboba &amp; Wood Standards Track [Page 29]</span></pre>
1621 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-30" ></span>
1622 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1623
1624
1625 FailOver(pcb);
1626 SetWatchdog();
1627 break ;
1628 case SUSPECT:
1629 pcb-&gt;status = DOWN;
1630 CloseConnection(pcb);
1631 SetWatchdog();
1632 break;
1633 case INITIAL:
1634 case DOWN:
1635 AttemptOpen(pcb);
1636 SetWatchdog();
1637 break;
1638 case REOPEN:
1639 if (!Pending) {
1640 SendWatchdog(pbc);
1641 SetWatchdog();
1642 Pending = TRUE;
1643 break;
1644 }
1645 if (NumDWA &lt; 0) {
1646 pcb-&gt;status = DOWN;
1647 CloseConnection(pcb);
1648 } else {
1649 NumDWA = -1;
1650 }
1651 SetWatchdog();
1652 break;
1653 default:
1654 error("Shouldn't be here!);
1655 break;
1656 }
1657 }
1658
1659 /*
1660 OnConnectionUp() is called whenever a connection comes up
1661 */
1662 OnConnectionUp(pcb)
1663 {
1664 switch (pcb-&gt;status){
1665 case INITIAL:
1666 pcb-&gt;status = OKAY;
1667 SetWatchdog();
1668 break;
1669 case DOWN:
1670 pcb-&gt;status = REOPEN;
1671 NumDWA = 0;
1672 SendWatchdog(pcb);
1673
1674
1675
1676 <span class="grey">Aboba &amp; Wood Standards Track [Page 30]</span></pre>
1677 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-31" ></span>
1678 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1679
1680
1681 SetWatchdog();
1682 Pending = TRUE;
1683 break;
1684 default:
1685 error("Shouldn't be here!);
1686 break;
1687 }
1688 }
1689
1690 /*
1691 OnConnectionDown() is called whenever a connection goes down
1692 */
1693 OnConnectionDown(pcb)
1694 {
1695 pcb-&gt;status = DOWN;
1696 CloseConnection();
1697 switch (pcb-&gt;status){
1698 case OKAY:
1699 Failover(pcb);
1700 SetWatchdog();
1701 break;
1702 case SUSPECT:
1703 case REOPEN:
1704 SetWatchdog();
1705 break;
1706 default:
1707 error("Shouldn't be here!);
1708 break;
1709 }
1710 }
1711
1712 /* Here is the state machine equivalent to the above code:
1713
1714 STATE Event Actions New State
1715 ===== ------ ------- ----------
1716 OKAY Receive DWA Pending = FALSE
1717 SetWatchdog() OKAY
1718 OKAY Receive non-DWA SetWatchdog() OKAY
1719 SUSPECT Receive DWA Pending = FALSE
1720 Failback()
1721 SetWatchdog() OKAY
1722 SUSPECT Receive non-DWA Failback()
1723 SetWatchdog() OKAY
1724 REOPEN Receive DWA &amp; Pending = FALSE
1725 NumDWA == 2 NumDWA++
1726 Failback() OKAY
1727 REOPEN Receive DWA &amp; Pending = FALSE
1728 NumDWA &lt; 2 NumDWA++ REOPEN
1729
1730
1731
1732 <span class="grey">Aboba &amp; Wood Standards Track [Page 31]</span></pre>
1733 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-32" ></span>
1734 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1735
1736
1737 STATE Event Actions New State
1738 ===== ------ ------- ----------
1739 REOPEN Receive non-DWA Throwaway() REOPEN
1740 INITIAL Receive DWA Pending = FALSE
1741 Throwaway() INITIAL
1742 INITIAL Receive non-DWA Throwaway() INITIAL
1743 DOWN Receive DWA Pending = FALSE
1744 Throwaway() DOWN
1745 DOWN Receive non-DWA Throwaway() DOWN
1746 OKAY Timer expires &amp; SendWatchdog()
1747 !Pending SetWatchdog()
1748 Pending = TRUE OKAY
1749 OKAY Timer expires &amp; Failover()
1750 Pending SetWatchdog() SUSPECT
1751 SUSPECT Timer expires CloseConnection()
1752 SetWatchdog() DOWN
1753 INITIAL Timer expires AttemptOpen()
1754 SetWatchdog() INITIAL
1755 DOWN Timer expires AttemptOpen()
1756 SetWatchdog() DOWN
1757 REOPEN Timer expires &amp; SendWatchdog()
1758 !Pending SetWatchdog()
1759 Pending = TRUE REOPEN
1760 REOPEN Timer expires &amp; CloseConnection()
1761 Pending &amp; SetWatchdog()
1762 NumDWA &lt; 0 DOWN
1763 REOPEN Timer expires &amp; NumDWA = -1
1764 Pending &amp; SetWatchdog()
1765 NumDWA &gt;= 0 REOPEN
1766 INITIAL Connection up SetWatchdog() OKAY
1767 DOWN Connection up NumDWA = 0
1768 SendWatchdog()
1769 SetWatchdog()
1770 Pending = TRUE REOPEN
1771 OKAY Connection down CloseConnection()
1772 Failover()
1773 SetWatchdog() DOWN
1774 SUSPECT Connection down CloseConnection()
1775 SetWatchdog() DOWN
1776 REOPEN Connection down CloseConnection()
1777 SetWatchdog() DOWN
1778 */
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788 <span class="grey">Aboba &amp; Wood Standards Track [Page 32]</span></pre>
1789 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-33" ></span>
1790 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1791
1792
1793 Appendix B - AAA Agents
1794
1795 As described in [<a href="./rfc2865" title="&quot;Remote Authentication Dial In User Service (RADIUS)&quot;">RFC2865</a>] and [<a href="./rfc2607" title="&quot;Proxy Chaining and Policy Implementation in Roaming&quot;">RFC2607</a>], AAA agents have become
1796 popular in order to support services such as roaming and shared use
1797 networks. Such agents are used both for
1798 authentication/authorization, as well as accounting [<a href="./rfc2975" title="&quot;Introduction to Accounting Management&quot;">RFC2975</a>].
1799
1800 AAA agents include:
1801
1802 Relays
1803 Proxies
1804 Re-directs
1805 Store and Forward proxies
1806 Transport layer proxies
1807
1808 The transport layer behavior of each of these agents is described
1809 below.
1810
1811 <span class="h3"><a class="selflink" id="appendix-B.1" href="#appendix-B.1">B.1</a> Relays and Proxies</span>
1812
1813 While the application-layer behavior of relays and proxies are
1814 different, at the transport layer the behavior is similar. In both
1815 cases, two connections are established: one from the AAA client (NAS)
1816 to the relay/proxy, and another from the relay/proxy to the AAA
1817 server. The relay/proxy does not respond to a client request until
1818 it receives a response from the server. Since the two connections
1819 are de-coupled, the end-to-end conversation between the client and
1820 server may not self clock.
1821
1822 Since AAA transport is typically application-driven, there is
1823 frequently not enough traffic to enable ACK piggybacking. As a
1824 result, the Nagle algorithm is rarely triggered, and delayed ACKs may
1825 comprise nearly half the traffic. Thus AAA protocols running over
1826 reliable transport will see packet traffic nearly double that
1827 experienced with UDP transport. Since ACK parameters (such as the
1828 value of the delayed ACK timer) are typically fixed by the TCP
1829 implementation and are not tunable by the application, there is
1830 little that can be done about this.
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844 <span class="grey">Aboba &amp; Wood Standards Track [Page 33]</span></pre>
1845 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-34" ></span>
1846 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1847
1848
1849 A typical trace of a conversation between a NAS, proxy and server is
1850 shown below:
1851
1852 Time NAS Relay/Proxy Server
1853 ------ --- ----------- ------
1854
1855 0 Request
1856 -------&gt;
1857 OTTnp + Tpr Request
1858 -------&gt;
1859
1860 OTTnp + TdA Delayed ACK
1861 &lt;-------
1862
1863 OTTnp + OTTps + Reply/ACK
1864 Tpr + Tsr &lt;-------
1865
1866 OTTnp + OTTps +
1867 Tpr + Tsr + Reply
1868 OTTsp + TpR &lt;-------
1869
1870 OTTnp + OTTps +
1871 Tpr + Tsr + Delayed ACK
1872 OTTsp + TdA -------&gt;
1873
1874 OTTnp + OTTps +
1875 OTTsp + OTTpn +
1876 Tpr + Tsr + Delayed ACK
1877 TpR + TdA -------&gt;
1878
1879 Key
1880 ---
1881 OTT = One-way Trip Time
1882 OTTnp = One-way trip time (NAS to Relay/Proxy)
1883 OTTpn = One-way trip time (Relay/Proxy to NAS)
1884 OTTps = One-way trip time (Relay/Proxy to Server)
1885 OTTsp = One-way trip time (Server to Relay/Proxy)
1886 TdA = Delayed ACK timer
1887 Tpr = Relay/Proxy request processing time
1888 TpR = Relay/Proxy reply processing time
1889 Tsr = Server request processing time
1890
1891 At time 0, the NAS sends a request to the relay/proxy. Ignoring the
1892 serialization time, the request arrives at the relay/proxy at time
1893 OTTnp, and the relay/proxy takes an additional Tpr in order to
1894 forward the request toward the home server. At time TdA after
1895
1896
1897
1898
1899
1900 <span class="grey">Aboba &amp; Wood Standards Track [Page 34]</span></pre>
1901 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-35" ></span>
1902 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1903
1904
1905 receiving the request, the relay/proxy sends a delayed ACK. The
1906 delayed ACK is sent, rather than being piggybacked on the reply, as
1907 long as TdA &lt; OTTps + OTTsp + Tpr + Tsr + TpR.
1908
1909 Typically Tpr &lt; TdA, so that the delayed ACK is sent after the
1910 relay/proxy forwards the request toward the server, but before the
1911 relay/proxy receives the reply from the server. However, depending
1912 on the TCP implementation on the relay/proxy and when the request is
1913 received, it is also possible for the delayed ACK to be sent prior to
1914 forwarding the request.
1915
1916 At time OTTnp + OTTps + Tpr, the server receives the request, and Tsr
1917 later, it generates the reply. Where Tsr &lt; TdA, the reply will
1918 contain a piggybacked ACK. However, depending on the server
1919 responsiveness and TCP implementation, the ACK and reply may be sent
1920 separately. This can occur, for example, where a slow database or
1921 storage system must be accessed prior to sending the reply.
1922
1923 At time OTTnp + OTTps + OTTsp + Tpr + Tsr the reply/ACK reaches the
1924 relay/proxy, which then takes TpR additional time to forward the
1925 reply to the NAS. At TdA after receiving the reply, the relay/proxy
1926 generates a delayed ACK. Typically TpR &lt; TdA so that the delayed ACK
1927 is sent to the server after the relay/proxy forwards the reply to the
1928 NAS. However, depending on the circumstances and the relay/proxy TCP
1929 implementation, the delayed ACK may be sent first.
1930
1931 As with a delayed ACK sent in response to a request, which may be
1932 piggybacked if the reply can be received quickly enough, piggybacking
1933 of the ACK sent in response to a reply from the server is only
1934 possible if additional request traffic is available. However, due to
1935 the high inter-packet spacings in typical AAA scenarios, this is
1936 unlikely unless the AAA protocol supports a reply ACK.
1937
1938 At time OTTnp + OTTps + OTTsp + OTTpn + Tpr + Tsr + TpR the NAS
1939 receives the reply. TdA later, a delayed ACK is generated.
1940
1941 <span class="h3"><a class="selflink" id="appendix-B.2" href="#appendix-B.2">B.2</a> Re-directs</span>
1942
1943 Re-directs operate by referring a NAS to the AAA server, enabling the
1944 NAS to talk to the AAA server directly. Since a direct transport
1945 connection is established, the end-to-end connection will self-clock.
1946
1947 With re-directs, delayed ACKs are less frequent than with
1948 application-layer proxies since the Re-direct and Server will
1949 typically piggyback replies with ACKs.
1950
1951
1952
1953
1954
1955
1956 <span class="grey">Aboba &amp; Wood Standards Track [Page 35]</span></pre>
1957 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-36" ></span>
1958 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
1959
1960
1961 The sequence of events is as follows:
1962
1963 Time NAS Re-direct Server
1964 ------ --- --------- ------
1965
1966 0 Request
1967 -------&gt;
1968 OTTnp + Tpr Redirect/ACK
1969 &lt;-------
1970
1971 OTTnp + Tpr + Request
1972 OTTpn + Tnr -------&gt;
1973
1974 OTTnp + OTTpn +
1975 Tpr + Tsr + Reply/ACK
1976 OTTns &lt;-------
1977
1978 OTTnp + OTTpn +
1979 OTTns + OTTsn +
1980 Tpr + Tsr + Delayed ACK
1981 TdA -------&gt;
1982
1983 Key
1984 ---
1985 OTT = One-way Trip Time
1986 OTTnp = One-way trip time (NAS to Re-direct)
1987 OTTpn = One-way trip time (Re-direct to NAS)
1988 OTTns = One-way trip time (NAS to Server)
1989 OTTsn = One-way trip time (Server to NAS)
1990 TdA = Delayed ACK timer
1991 Tpr = Re-direct processing time
1992 Tnr = NAS re-direct processing time
1993 Tsr = Server request processing time
1994
1995 <span class="h3"><a class="selflink" id="appendix-B.3" href="#appendix-B.3">B.3</a> Store and Forward Proxies</span>
1996
1997 With a store and forward proxy, the proxy may send a reply to the NAS
1998 prior to forwarding the request to the server. While store and
1999 forward proxies are most frequently deployed for accounting
2000 [<a href="./rfc2975" title="&quot;Introduction to Accounting Management&quot;">RFC2975</a>], they also can be used to implement
2001 authentication/authorization policy, as described in [<a href="./rfc2607" title="&quot;Proxy Chaining and Policy Implementation in Roaming&quot;">RFC2607</a>].
2002
2003 As noted in [<a href="./rfc2975" title="&quot;Introduction to Accounting Management&quot;">RFC2975</a>], store and forward proxies can have a negative
2004 effect on accounting reliability. By sending a reply to the NAS
2005 without receiving one from the accounting server, store and forward
2006 proxies fool the NAS into thinking that the accounting request had
2007 been accepted by the accounting server when this is not the case. As
2008 a result, the NAS can delete the accounting packet from non-volatile
2009
2010
2011
2012 <span class="grey">Aboba &amp; Wood Standards Track [Page 36]</span></pre>
2013 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-37" ></span>
2014 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
2015
2016
2017 storage before it has been accepted by the accounting server. That
2018 leaves the proxy responsible for delivering accounting packets. If
2019 the proxy involves moving parts (e.g. a disk drive) while the NAS
2020 does not, overall system reliability can be reduced. As a result,
2021 store and forward proxies SHOULD NOT be used.
2022
2023 The sequence of events is as follows:
2024
2025 Time NAS Proxy Server
2026 ------ --- ----- ------
2027
2028 0 Request
2029 -------&gt;
2030 OTTnp + TpR Reply/ACK
2031 &lt;-------
2032
2033 OTTnp + Tpr Request
2034 -------&gt;
2035
2036 OTTnp + OTTph + Reply/ACK
2037 Tpr + Tsr &lt;-------
2038
2039 OTTnp + OTTph +
2040 Tpr + Tsr + Reply
2041 OTThp + TpR &lt;-------
2042
2043 OTTnp + OTTph +
2044 Tpr + Tsr + Delayed ACK
2045 OTThp + TdA -------&gt;
2046
2047 OTTnp + OTTph +
2048 OTThp + OTTpn +
2049 Tpr + Tsr + Delayed ACK
2050 TpR + TdA -------&gt;
2051
2052 Key
2053 ---
2054 OTT = One-way Trip Time
2055 OTTnp = One-way trip time (NAS to Proxy)
2056 OTTpn = One-way trip time (Proxy to NAS)
2057 OTTph = One-way trip time (Proxy to Home server)
2058 OTThp = One-way trip time (Home Server to Proxy)
2059 TdA = Delayed ACK timer
2060 Tpr = Proxy request processing time
2061 TpR = Proxy reply processing time
2062 Tsr = Server request processing time
2063
2064
2065
2066
2067
2068 <span class="grey">Aboba &amp; Wood Standards Track [Page 37]</span></pre>
2069 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-38" ></span>
2070 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
2071
2072
2073 <span class="h3"><a class="selflink" id="appendix-B.4" href="#appendix-B.4">B.4</a> Transport Layer Proxies</span>
2074
2075 In addition to acting as proxies at the application layer, transport
2076 layer proxies forward transport ACKs between the AAA client and
2077 server. This splices together the client-proxy and proxy-server
2078 connections into a single connection that behaves as though it
2079 operates end-to-end, exhibiting self-clocking. However, since
2080 transport proxies operate at the transport layer, they cannot be
2081 implemented purely as applications and they are rarely deployed.
2082
2083 With a transport proxy, the sequence of events is as follows:
2084
2085 Time NAS Proxy Home Server
2086 ------ --- ----- -----------
2087
2088 0 Request
2089 -------&gt;
2090 OTTnp + Tpr Request
2091 -------&gt;
2092
2093 OTTnp + OTTph + Reply/ACK
2094 Tpr + Tsr &lt;-------
2095
2096 OTTnp + OTTph +
2097 Tpr + Tsr + Reply/ACK
2098 OTThp + TpR &lt;-------
2099
2100 OTTnp + OTTph +
2101 OTThp + OTTpn +
2102 Tpr + Tsr + Delayed ACK
2103 TpR + TdA -------&gt;
2104
2105
2106 OTTnp + OTTph +
2107 OTThp + OTTpn +
2108 Tpr + Tsr + Delayed ACK
2109 TpR + TpD -------&gt;
2110
2111 Key
2112 ---
2113 OTT = One-way Trip Time
2114 OTTnp = One-way trip time (NAS to Proxy)
2115 OTTpn = One-way trip time (Proxy to NAS)
2116 OTTph = One-way trip time (Proxy to Home server)
2117 OTThp = One-way trip time (Home Server to Proxy)
2118 TdA = Delayed ACK timer
2119 Tpr = Proxy request processing time
2120 TpR = Proxy reply processing time
2121
2122
2123
2124 <span class="grey">Aboba &amp; Wood Standards Track [Page 38]</span></pre>
2125 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-39" ></span>
2126 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
2127
2128
2129 Tsr = Server request processing time
2130 TpD = Proxy delayed ack processing time
2131
2132 Intellectual Property Statement
2133
2134 The IETF takes no position regarding the validity or scope of any
2135 intellectual property or other rights that might be claimed to
2136 pertain to the implementation or use of the technology described in
2137 this document or the extent to which any license under such rights
2138 might or might not be available; neither does it represent that it
2139 has made any effort to identify any such rights. Information on the
2140 IETF's procedures with respect to rights in standards-track and
2141 standards-related documentation can be found in <a href="https://www.rfc-editor.org/bcp/bcp11">BCP-11</a>. Copies of
2142 claims of rights made available for publication and any assurances of
2143 licenses to be made available, or the result of an attempt made to
2144 obtain a general license or permission for the use of such
2145 proprietary rights by implementors or users of this specification can
2146 be obtained from the IETF Secretariat.
2147
2148 The IETF invites any interested party to bring to its attention any
2149 copyrights, patents or patent applications, or other proprietary
2150 rights which may cover technology that may be required to practice
2151 this standard. Please address the information to the IETF Executive
2152 Director.
2153
2154 Acknowledgments
2155
2156 Thanks to Allison Mankin of AT&amp;T, Barney Wolff of Databus, Steve Rich
2157 of Cisco, Randy Bush of AT&amp;T, Bo Landarv of IP Unplugged, Jari Arkko
2158 of Ericsson, and Pat Calhoun of Blackstorm Networks for fruitful
2159 discussions relating to AAA transport.
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180 <span class="grey">Aboba &amp; Wood Standards Track [Page 39]</span></pre>
2181 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-40" ></span>
2182 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
2183
2184
2185 Authors' Addresses
2186
2187 Bernard Aboba
2188 Microsoft Corporation
2189 One Microsoft Way
2190 Redmond, WA 98052
2191
2192 Phone: +1 425 706 6605
2193 Fax: +1 425 936 7329
2194 EMail: [email protected]
2195
2196
2197 Jonathan Wood
2198 Sun Microsystems, Inc.
2199 901 San Antonio Road
2200 Palo Alto, CA 94303
2201
2202 EMail: [email protected]
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236 <span class="grey">Aboba &amp; Wood Standards Track [Page 40]</span></pre>
2237 <hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-41" ></span>
2238 <span class="grey"><a href="./rfc3539">RFC 3539</a> AAA Transport Profile June 2003</span>
2239
2240
2241 Full Copyright Statement
2242
2243 Copyright (C) The Internet Society (2003). All Rights Reserved.
2244
2245 This document and translations of it may be copied and furnished to
2246 others, and derivative works that comment on or otherwise explain it
2247 or assist in its implementation may be prepared, copied, published
2248 and distributed, in whole or in part, without restriction of any
2249 kind, provided that the above copyright notice and this paragraph are
2250 included on all such copies and derivative works. However, this
2251 document itself may not be modified in any way, such as by removing
2252 the copyright notice or references to the Internet Society or other
2253 Internet organizations, except as needed for the purpose of
2254 developing Internet standards in which case the procedures for
2255 copyrights defined in the Internet Standards process must be
2256 followed, or as required to translate it into languages other than
2257 English.
2258
2259 The limited permissions granted above are perpetual and will not be
2260 revoked by the Internet Society or its successors or assigns.
2261
2262 This document and the information contained herein is provided on an
2263 "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
2264 TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
2265 BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
2266 HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
2267 MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
2268
2269 Acknowledgement
2270
2271 Funding for the RFC Editor function is currently provided by the
2272 Internet Society.
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292 Aboba &amp; Wood Standards Track [Page 41]
2293 </pre>