idnits 2.17.1 draft-ietf-issll-is802-framework-06.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- ** Looks like you're using RFC 2026 boilerplate. This must be updated to follow RFC 3978/3979, as updated by RFC 4748. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- ** Missing expiration date. The document expiration date should appear on the first and last page. ** The document is more than 15 pages and seems to lack a Table of Contents. == No 'Intended status' indicated for this document; assuming Proposed Standard == It seems as if not all pages are separated by form feeds - found 0 form feeds but 48 pages Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- ** The document seems to lack an IANA Considerations section. (See Section 2.2 of https://www.ietf.org/id-info/checklist for how to handle the case when there are no actions for IANA.) ** The document seems to lack separate sections for Informative/Normative References. All references will be assumed normative when checking for downward references. ** The abstract seems to contain references ([13], [14]), which it shouldn't. Please replace those with straight textual mentions of the documents in question. Miscellaneous warnings: ---------------------------------------------------------------------------- == Line 1741 has weird spacing: '... 1.2 ms unb...' == Line 1742 has weird spacing: '... 120 us unb...' == Line 1743 has weird spacing: '... 12 us unb...' == Line 1772 has weird spacing: '... 1.2 ms unb...' == Line 1773 has weird spacing: '... 120 us unb...' == (1 more instance...) -- The document seems to lack a disclaimer for pre-RFC5378 work, but may have content which was first submitted before 10 November 2008. If you have contacted all the original authors and they are all willing to grant the BCP78 rights to the IETF Trust, then this is fine, and you can ignore this comment. If not, you may need to add the pre-RFC5378 disclaimer. (See the Legal Provisions document at https://trustee.ietf.org/license-info for more information.) -- The document date (May 1999) is 9110 days in the past. Is this intentional? Checking references for intended status: Proposed Standard ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) == Missing Reference: '16' is mentioned on line 432, but not defined == Unused Reference: '11' is defined on line 2013, but no explicit reference was found in the text == Unused Reference: '12' is defined on line 2016, but no explicit reference was found in the text == Unused Reference: '17' is defined on line 2043, but no explicit reference was found in the text == Unused Reference: '18' is defined on line 2046, but no explicit reference was found in the text == Unused Reference: '20' is defined on line 2054, but no explicit reference was found in the text -- Possible downref: Non-RFC (?) normative reference: ref. '1' -- Possible downref: Non-RFC (?) normative reference: ref. '2' -- Possible downref: Non-RFC (?) normative reference: ref. '3' -- Possible downref: Non-RFC (?) normative reference: ref. '4' ** Downref: Normative reference to an Informational RFC: RFC 1633 (ref. '8') ** Downref: Normative reference to an Informational RFC: RFC 2216 (ref. '10') ** Downref: Normative reference to an Historic RFC: RFC 1819 (ref. '12') -- Possible downref: Non-RFC (?) normative reference: ref. '13' -- Possible downref: Non-RFC (?) normative reference: ref. '14' -- Possible downref: Non-RFC (?) normative reference: ref. '15' -- Possible downref: Non-RFC (?) normative reference: ref. '18' -- Possible downref: Non-RFC (?) normative reference: ref. '19' -- Possible downref: Non-RFC (?) normative reference: ref. '20' Summary: 9 errors (**), 0 flaws (~~), 14 warnings (==), 12 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 1 Internet Engineering Task Force Anoop Ghanwani 2 INTERNET DRAFT (Nortel Networks) 3 J. Wayne Pace 4 (IBM) 5 Vijay Srinivasan 6 (Torrent Networking Technologies) 7 Andrew Smith 8 (Extreme Networks) 9 Mick Seaman 10 (3Com) 11 May 1999 13 A Framework for Providing Integrated Services 14 Over Shared and Switched IEEE 802 LAN Technologies 16 draft-ietf-issll-is802-framework-06.txt 18 Status of This Memo 20 This document is an Internet-Draft and is in full conformance with 21 all provisions of Section 10 of RFC2026. 23 Internet-Drafts are working documents of the Internet Engineering 24 Task Force (IETF), its areas, and its working groups. Note that 25 other groups may also distribute working documents as Internet- 26 Drafts. 28 Internet-Drafts are draft documents valid for a maximum of six months 29 and may be updated, replaced, or obsoleted by other documents at 30 any time. It is inappropriate to use Internet-Drafts as reference 31 material or to cite them other than as "work in progress." 33 The list of current Internet-Drafts can be accessed at 34 http://www.ietf.org/ietf/1id-abstracts.txt 36 The list of Internet-Draft Shadow Directories can be accessed at 37 http://www.ietf.org/shadow.html. 39 Abstract 41 This memo describes a framework for supporting IETF Integrated 42 Services on shared and switched LAN infrastructure. It includes 43 background material on the capabilities of IEEE 802 like networks 44 with regard to parameters that affect Integrated Services such as 45 access latency, delay variation and queueing support in LAN switches. 46 It discusses aspects of IETF's Integrated Services model that cannot 47 easily be accommodated in different LAN environments. It outlines 48 a functional model for supporting the Resource Reservation Protocol 49 (RSVP) in such LAN environments. Details of extensions to RSVP for 50 use over LANs are described in an accompanying memo [14]. Mappings 51 of the various Integrated Services onto IEEE 802 LANs are described 52 in another memo [13]. 54 Contents 56 Status of This Memo i 58 Abstract ii 60 1. Introduction 1 62 2. Document Outline 1 64 3. Definitions 2 66 4. Frame Forwarding in IEEE 802 Networks 3 67 4.1. General IEEE 802 Service Model . . . . . . . . . . . . . 3 68 4.2. Ethernet/IEEE 802.3 . . . . . . . . . . . . . . . . . . . 5 69 4.3. Token Ring/IEEE 802.5 . . . . . . . . . . . . . . . . . . 6 70 4.4. Fiber Distributed Data Interface . . . . . . . . . . . . 7 71 4.5. Demand Priority/IEEE 802.12 . . . . . . . . . . . . . . . 8 73 5. Requirements and Goals 9 74 5.1. Requirements . . . . . . . . . . . . . . . . . . . . . . 9 75 5.2. Goals . . . . . . . . . . . . . . . . . . . . . . . . . . 11 76 5.3. Non-goals . . . . . . . . . . . . . . . . . . . . . . . . 12 77 5.4. Assumptions . . . . . . . . . . . . . . . . . . . . . . . 12 79 6. Basic Architecture 13 80 6.1. Components . . . . . . . . . . . . . . . . . . . . . . . 13 81 6.1.1. Requester Module . . . . . . . . . . . . . . . . 13 82 6.1.2. Bandwidth Allocator . . . . . . . . . . . . . . . 14 83 6.1.3. Communication Protocols . . . . . . . . . . . . . 14 84 6.2. Centralized vs. Distributed Implementations . . . . . . 15 86 7. Model of the Bandwidth Manager in a Network 17 87 7.1. End Station Model . . . . . . . . . . . . . . . . . . . . 17 88 7.1.1. Layer 3 Client Model . . . . . . . . . . . . . . 17 89 7.1.2. Requests to Layer 2 ISSLL . . . . . . . . . . . . 17 90 7.1.3. At the Layer 3 Sender . . . . . . . . . . . . . . 18 91 7.1.4. At the Layer 3 Receiver . . . . . . . . . . . . . 19 92 7.2. Switch Model . . . . . . . . . . . . . . . . . . . . . . 21 93 7.2.1. Centralized Bandwidth Allocator . . . . . . . . . 21 94 7.2.2. Distributed Bandwidth Allocator . . . . . . . . . 21 95 7.3. Admission Control . . . . . . . . . . . . . . . . . . . . 22 96 7.4. QoS Signaling . . . . . . . . . . . . . . . . . . . . . . 24 97 7.4.1. Client Service Definitions . . . . . . . . . . . 24 98 7.4.2. Switch Service Definitions . . . . . . . . . . . 25 100 8. Implementation Issues 27 101 8.1. Switch Characteristics . . . . . . . . . . . . . . . . . 27 102 8.2. Queueing . . . . . . . . . . . . . . . . . . . . . . . . 28 103 8.3. Mapping of Services to Link Level Priority . . . . . . . 29 104 8.4. Re-mapping of Non-conforming Aggregated Flows . . . . . . 29 105 8.5. Override of Incoming User Priority . . . . . . . . . . . 30 106 8.6. Different Reservation Styles . . . . . . . . . . . . . . 30 107 8.7. Receiver Heterogeneity . . . . . . . . . . . . . . . . . 31 109 9. Network Topology Scenarios 34 110 9.1. Full Duplex Switched Networks . . . . . . . . . . . . . . 35 111 9.2. Shared Media Ethernet Networks . . . . . . . . . . . . . 35 112 9.3. Half Duplex Switched Ethernet Networks . . . . . . . . . 36 113 9.4. Half Duplex Switched and Shared Token Ring Networks . . . 37 114 9.5. Half Duplex and Shared Demand Priority Networks . . . . . 38 116 10. Justification 41 118 11. Summary 41 119 1. Introduction 121 The Internet has traditionally provided support for best effort 122 traffic only. However, with the recent advances in link layer 123 technology, and with numerous emerging real time applications such 124 as video conferencing and Internet telephony, there has been much 125 interest for developing mechanisms which enable real time services 126 over the Internet. A framework for meeting these new requirements 127 was set out in RFC 1633 [8] and this has driven the specification of 128 various classes of network service by the Integrated Services working 129 group of the IETF, such as Controlled Load and Guaranteed Service 130 [6,7]. Each of these service classes is designed to provide certain 131 Quality of Service (QoS) to traffic conforming to a specified set 132 of parameters. Applications are expected to choose one of these 133 classes according to their QoS requirements. One mechanism for end 134 stations to utilize such services in an IP network is provided by 135 a QoS signaling protocol, the Resource Reservation Protocol (RSVP) 136 [5] developed by the RSVP working group of the IETF. The IEEE under 137 its Project 802 has defined standards for many different local area 138 network technologies. These all typically offer the same MAC layer 139 datagram service [1] to higher layer protocols such as IP although 140 they often provide different dynamic behavior characteristics -- it 141 is these that are important when considering their ability to support 142 real time services. Later in this memo we describe some of the 143 relevant characteristics of the different MAC layer LAN technologies. 144 In addition, IEEE 802 has defined standards for bridging multiple LAN 145 segments together using devices known as "MAC Bridges" or "Switches" 146 [2]. Recent work has also defined traffic classes, multicast 147 filtering, and virtual LAN capabilities for these devices [3,4]. 148 Such LAN technologies often constitute the last hop(s) between users 149 and the Internet as well as being a primary building block for entire 150 campus networks. It is therefore necessary to provide standardized 151 mechanisms for using these technologies to support end-to-end real 152 time services. In order to do this, there must be some mechanism 153 for resource management at the data link layer. Resource management 154 in this context encompasses the functions of admission control, 155 scheduling, traffic policing, etc. The ISSLL (Integrated Services 156 over Specific Link Layers) working group in the IETF was chartered 157 with the purpose of exploring and standardizing such mechanisms for 158 various link layer technologies. 160 2. Document Outline 162 This document is concerned with specifying a framework for providing 163 Integrated Services over shared and switched LAN technologies such 164 as Ethernet/IEEE 802.3, Token Ring/IEEE 802.5, FDDI, etc. We begin 165 in Section 4 with a discussion of the capabilities of various IEEE 166 802 MAC layer technologies. Section 5 lists the requirements and 167 goals for a mechanism capable of providing Integrated Services in 168 a LAN. The resource management functions outlined in Section 5 are 169 provided by an entity referred to as a Bandwidth Manager (BM). The 170 architectural model of the the BM is described in Section 6 and its 171 various components are discussed in Section 7. Some implementation 172 issues with respect to link layer support for Integrated Services are 173 examined in Section 8. Section 9 discusses a taxonomy of topologies 174 for the LAN technologies under consideration with an emphasis 175 on the capabilities of each which can be leveraged for enabling 176 Integrated Services. This framework makes no assumptions about the 177 topology at the link layer. The framework is intended to be as 178 exhaustive as possible; this means that it is possible that all the 179 functions discussed may not be supportable by a particular topology 180 or technology, but this should not preclude the usage of this model 181 for it. 183 3. Definitions 185 The following is a list of terms used in this and other ISSLL 186 documents. 188 - Link Layer or Layer 2 or L2: Data link layer technologies such 189 as Ethernet/IEEE 802.3 and Token Ring/IEEE 802.5 are referred to 190 as Layer 2 or L2. 192 - Link Layer Domain or Layer 2 Domain or L2 Domain: Refers to a 193 set of nodes and links interconnected without passing through a 194 L3 forwarding function. One or more IP subnets can be overlaid 195 on a L2 domain. 197 - Layer 2 or L2 Devices: Devices that only implement Layer 2 198 functionality as Layer 2 or L2 devices. These include IEEE 199 802.1D [2] bridges or switches. 201 - Internetwork Layer or Layer 3 or L3: Refers to Layer 3 of the 202 ISO OSI model. This memo is primarily concerned with networks 203 that use the Internet Protocol (IP) at this layer. 205 - Layer 3 Device or L3 Device or End Station: These include hosts 206 and routers that use L3 and higher layer protocols or application 207 programs that need to make resource reservations. 209 - Segment: A physical L2 segment that is shared by one or more 210 senders. Examples of segments include: (a) a shared Ethernet or 211 Token Ring wire resolving contention for media access using CSMA 212 or token passing; (b) a half duplex link between two stations or 213 switches; (c) one direction of a switched full duplex link. 215 - Managed Segment: A managed segment is a segment with a DSBM 216 present and responsible for exercising admission control over 217 requests for resource reservation. A managed segment includes 218 those interconnected parts of a shared LAN that are not separated 219 by DSBMs. 221 - Traffic Class: Refers to an aggregation of data flows which are 222 given similar service within a switched network. 224 - Subnet: Used in this memo to indicate a group of L3 devices 225 sharing a common L3 network address prefix along with the set of 226 segments making up the L2 domain in which they are located. 228 - Bridge/Switch: A Layer 2 forwarding device as defined by IEEE 229 802.1D [2]. The terms bridge and switch are used synonymously in 230 this memo. 232 4. Frame Forwarding in IEEE 802 Networks 234 4.1. General IEEE 802 Service Model 236 The user_priority is a value associated with the transmission 237 and reception of all frames in the IEEE 802 service model. It 238 is supplied by the sender that is using the MAC service and is 239 provided along with the data to a receiver using the MAC service. 240 It may or may not be actually carried over the network. Token 241 Ring/IEEE 802.5 carries this value encoded in its FC octet while 242 basic Ethernet/IEEE 802.3 does not carry it. IEEE 802.12 may or 243 may not carry it depending on the frame format in use. When the 244 frame format in use is IEEE 802.5, the user_priority is carried 245 explicitly. When IEEE 802.3 frame format is used, only the two 246 levels of priority (high/low) that are used to determine access 247 priority can be recovered. This is based on the value of priority 248 encoded in the start delimiter of the IEEE 802.3 frame. 250 IEEE 802.1D [3] (1) defines a consistent way carry this value over a 251 bridged network consisting of Ethernet, Token Ring, Demand Priority, 253 ---------------------------- 254 1. The original IEEE 802.1D standard [2] contains the specifications for 255 the operation of MAC bridges. This has recently been extended to 256 include support for traffic classes and dynamic multicast filtering 257 [3]. In this document, the reader should be aware that references 258 to the IEEE 802.1D standard refer to [3], unless explicitly noted 259 otherwise. 261 FDDI or other MAC layer media using an extended frame format. The 262 usage of user_priority is summarized below. We refer the interested 263 reader to the IEEE 802.1D specification for further information. 265 If the user_priority is carried explicitly in packets, its utility is 266 as a simple label enabling packets within a data stream in different 267 classes to be discriminated easily by downstream nodes without having 268 to parse the packet in more detail. 270 Apart from making the job of desktop or wiring closet switches 271 easier, an explicit field means they do not have to change hardware 272 or software as the rules for classifying packets evolve; e.g. 273 based on new protocols or new policies. More sophisticated Layer 274 3 switches, perhaps deployed in the core of a network, may be able 275 to provide added value by performing packet classification more 276 accurately and, hence, utilizing network resources more efficiently 277 and providing better isolation between flows. This appears to be 278 a good economic choice since there are likely to be very many more 279 desktop/wiring closet switches in a network than switches requiring 280 Layer 3 functionality. 282 The IEEE 802 specifications make no assumptions about how 283 user_priority is to be used by end stations or by the network. 284 Although IEEE 802.1D defines static priority queueing as the default 285 mode of operation of switches that implement multiple queues, the 286 user_priority is really a priority only in a loose sense since it 287 depends on the number of traffic classes actually implemented by a 288 switch. The user_priority is defined as a 3 bit quantity with a 289 value of 7 representing the highest priority and a value of 0 as 290 the lowest. The general switch algorithm is as follows. Packets 291 are queued within a particular traffic class based on the received 292 user_priority, the value of which is either obtained directly from 293 the packet if an IEEE P802.1Q header or IEEE 802.5 network is used, 294 or is assigned according to some local policy. The queue is selected 295 based on a mapping from user_priority (0 through 7) onto the number 296 of available traffic classes. A switch may implement one or more 297 traffic classes. The advertised IntServ parameters and the switch's 298 admission control behavior may be used to determine the mapping from 299 user_priority to traffic classes within the switch. A switch is 300 not precluded from implementing other scheduling algorithms such as 301 weighted fair queueing and round robin. 303 IEEE 802.1D makes no recommendations about how a sender should 304 select the value for user_priority. One of the primary purposes of 305 this document is to propose such usage rules, and to discuss the 306 communication of the semantics of these values between switches and 307 end stations. In the remainder of this document we use the term 308 traffic class synonymously with user_priority. 310 4.2. Ethernet/IEEE 802.3 312 There is no explicit traffic class or user_priority field carried in 313 Ethernet packets. This means that user_priority must be regenerated 314 at a downstream receiver or switch according to some defaults or by 315 parsing further into higher layer protocol fields in the packet. 316 Alternatively, IEEE P802.1Q encapsulation [4] may be used which 317 provides an explicit user_priority field on top of the basic MAC 318 frame format. 320 For the different IP packet encapsulations used over Ethernet/IEEE 321 802.3, it will be necessary to adjust any admission control 322 calculations according to the framing and padding requirements. 324 Table 1: Ethernet encapsulations 326 --------------------------------------------------------------- 327 Encapsulation Framing Overhead IP MTU 328 bytes/pkt bytes 329 --------------------------------------------------------------- 330 IP EtherType (ip_len<=46 bytes) 64-ip_len 1500 331 (1500>=ip_len>=46 bytes) 18 1500 333 IP EtherType over 802.1D/Q (ip_len<=42) 64-ip_len 1500* 334 (1500>=ip_len>=42 bytes) 22 1500* 336 IP EtherType over LLC/SNAP (ip_len<=40) 64-ip_len 1492 337 (1500>=ip_len>=40 bytes) 24 1492 338 --------------------------------------------------------------- 340 *Note that the draft IEEE P802.1Q specification exceeds the current 341 IEEE 802.3 maximum packet length values by 4 bytes. The change of 342 maximum MTU size for IEEE P802.1Q frames is being accommodated by 343 IEEE P802.3ac. 345 4.3. Token Ring/IEEE 802.5 347 The Token Ring standard [6] provides a priority mechanism that can 348 be used to control both the queueing of packets for transmission and 349 the access of packets to the shared media. The priority mechanisms 350 are implemented using bits within the Access Control (AC) and the 351 Frame Control (FC) fields of a LLC frame. The first three bits of 352 the AC field, the Token Priority bits, together with the last three 353 bits of the AC field, the Reservation bits, regulate which stations 354 get access to the ring. The last three bits of the FC field of an 355 LLC frame, the User Priority bits, are obtained from the higher layer 356 in the user_priority parameter when it requests transmission of a 357 packet. This parameter also establishes the Access Priority used 358 by the MAC. The user_priority value is conveyed end-to-end by the 359 User Priority bits in the FC field and is typically preserved through 360 Token Ring bridges of all types. In all cases, 0 is the lowest 361 priority. 363 Token Ring also uses a concept of Reserved Priority which relates to 364 the value of priority which a station uses to reserve the token for 365 the next transmission on the ring. When a free token is circulating, 366 only a station having an Access Priority greater than or equal to the 367 Reserved Priority in the token will be allowed to seize the token for 368 transmission. Readers are referred to [14] for further discussion of 369 this topic. 371 A Token Ring station is theoretically capable of separately queueing 372 each of the eight levels of requested user_priority and then 373 transmitting frames in order of priority. A station sets Reservation 374 bits according to the user_priority of frames that are queued 375 for transmission in the highest priority queue. This allows the 376 access mechanism to ensure that the frame with the highest priority 377 throughout the entire ring will be transmitted before any lower 378 priority frame. Annex I to the IEEE 802.5 Token Ring standard 379 recommends that stations send/relay frames as follows. 381 To reduce frame jitter associated with high priority traffic, the 382 annex also recommends that only one frame be transmitted per token 383 and that the maximum information field size be 4399 octets whenever 384 delay sensitive traffic is traversing the ring. Most existing 385 implementations of Token Ring bridges forward all LLC frames with 386 a default access priority of 4. Annex I recommends that bridges 387 forward LLC frames that have a user_priority greater than 4 with 388 a reservation equal to the user_priority (although the draft IEEE 389 802.1D [3] permits network management override this behavior). The 390 capabilities provided by the Token Ring architecture, such User 391 Priority and Reserved Priority, can provide effective support for 392 Integrated Services flows that require QoS guarantees. 394 Table 2: Recommended use of Token Ring User Priority 396 ------------------------------------- 397 Application User Priority 398 ------------------------------------- 399 Non-time-critical data 0 400 - 1 401 - 2 402 - 3 403 LAN management 4 404 Time-sensitive data 5 405 Real-time-critical data 6 406 MAC frames 7 407 ------------------------------------- 409 For the different IP packet encapsulations used over Token Ring/IEEE 410 802.5, it will be necessary to adjust any admission control 411 calculations according to the framing requirements as shown in Table 412 3. 414 Table 3: Token Ring encapsulations 416 --------------------------------------------------------------- 417 Encapsulation Framing Overhead IP MTU 418 bytes/pkt bytes 419 --------------------------------------------------------------- 420 IP EtherType over 802.1D/Q 29 4370* 421 IP EtherType over LLC/SNAP 25 4370* 422 --------------------------------------------------------------- 424 *The suggested MTU from RFC 1042 [13] is 4464 bytes but there are 425 issues related to discovering what the maximum supported MTU between 426 any two points both within and between Token Ring subnets. The MTU 427 reported here is consistent with the IEEE 802.5 Annex I 428 recommendation. 430 4.4. Fiber Distributed Data Interface 432 The Fiber Distributed Data Interface (FDDI) standard [16] provides 433 a priority mechanism that can be used to control both the queueing 434 of packets for transmission and the access of packets to the shared 435 media. The priority mechanisms are implemented using similar 436 mechanisms to Token Ring described above. The standard also makes 437 provision for "Synchronous" data traffic with strict media access and 438 delay guarantees. This mode of operation is not discussed further 439 here and represents area within the scope of the ISSLL working group 440 that requires further work. In the remainder of this document, for 441 the discussion of QoS mechanisms, FDDI is treated as a 100 Mbps Token 442 Ring technology using a service interface compatible with IEEE 802 443 networks. 445 4.5. Demand Priority/IEEE 802.12 447 IEEE 802.12 [19] is a standard for a shared 100 Mbps LAN. Data 448 packets are transmitted using either the IEEE 802.3 or IEEE 802.5 449 frame format. The MAC protocol is called Demand Priority. Its main 450 characteristics with respect to QoS are the support of two service 451 priority levels, normal priority and high priority, and the order of 452 service for each of these. Data packets from all network nodes (end 453 hosts and bridges/switches) are served using a simple round robin 454 algorithm. 456 If the IEEE 802.3 frame format is used for data transmission then 457 the user_priority is encoded in the starting delimiter of the IEEE 458 802.12 data packet. If the IEEE 802.5 frame format is used then the 459 user_priority is additionally encoded in the YYY bits of the FC field 460 in the IEEE 802.5 packet header (see also Section 4.3). Furthermore, 461 the IEEE P802.1Q encapsulation with its own user_priority field may 462 also be applied in IEEE 802.12 networks. In all cases, switches are 463 able to recover any user_priority supplied by a sender. 465 The same rules apply for IEEE 802.12 user_priority mapping in a 466 bridge as with other media types. The only additional information 467 is that normal priority is used by default for user_priority values 468 0 through 4 inclusive, and high priority is used for user_priority 469 levels 5 through 7. This ensures that the default Token Ring 470 user_priority level of 4 for IEEE 802.5 bridges is mapped to normal 471 priority on IEEE 802.12 segments. 473 The medium access in IEEE 802.12 LANs is deterministic. The Demand 474 Priority mechanism ensures that, once the normal priority service 475 has been preempted, all high priority packets have strict priority 476 over packets with normal priority. In the abnormal situation that 477 a normal priority packet has been waiting at the head of line of a 478 MAC transmit queue for a time period longer than PACKET_PROMOTION 479 (200 - 300 ms) [19], its priority is automatically promoted to 480 high priority. Thus, even normal priority packets have a maximum 481 guaranteed access time to the medium. 483 Integrated Services can be built on top of the IEEE 802.12 medium 484 access mechanism. When combined with admission control and bandwidth 485 enforcement mechanisms, delay guarantees as required for a Guaranteed 486 Service can be provided without any changes to the existing IEEE 487 802.12 MAC protocol. 489 Since the IEEE 802.12 standard supports the IEEE 802.3 and IEEE 802.5 490 frame formats, the same framing overhead as reported in Sections 4.2 491 and 4.3 must be considered in the admission control computations for 492 IEEE 802.12 links. 494 5. Requirements and Goals 496 This section discusses the requirements and goals which should drive 497 the design of an architecture for supporting Integrated Services over 498 LAN technologies. The requirements refer to functions and features 499 which must be supported, while goals refer to functions and features 500 which are desirable, but are not an absolute necessity. Many of the 501 requirements and goals are driven by the functionality supported by 502 Integrated Services and RSVP. 504 5.1. Requirements 506 - Resource Reservation: The mechanism must be capable of reserving 507 resources on a single segment or multiple segments and at 508 bridges/switches connecting them. It must be able to provide 509 reservations for both unicast and multicast sessions. It should 510 be possible to change the level of reservation while the session 511 is in progress. 513 - Admission Control: The mechanism must be able to estimate 514 the level of resources necessary to meet the QoS requested by 515 the session in order to decide whether or not the session can 516 be admitted. For the purpose of management, it is useful to 517 provide the ability to respond to queries about availability of 518 resources. It must be able to make admission control decisions 519 for different types of services such as Guaranteed Service, 520 Controlled Load, etc. 522 - Flow Separation and Scheduling: It is necessary to provide a 523 mechanism for traffic flow separation so that real time flows can 524 be given preferential treatment over best effort flows. Packets 525 of real time flows can then be isolated and scheduled according 526 to their service requirements. 528 - Policing/Shaping: Traffic must be shaped and/or policed by 529 end stations (workstations, routers) to ensure conformance to 530 negotiated traffic parameters. Shaping is the recommended 531 behavior for traffic sources. A router initiating an ISSLL 532 session must have implemented traffic control mechanisms 533 according to the IntServ requirements which would ensure that 534 all flows sent by the router are in conformance. The ISSLL 535 mechanisms at the link layer rely heavily on the correct 536 implementation of policing/shaping mechanisms at higher layers by 537 devices capable of doing so. This is necessary because bridges 538 and switches are not typically capable of maintaining per flow 539 state which would be required to check flows for conformance. 540 Policing is left as an option for bridges and switches, which if 541 implemented, may be used to enforce tighter control over traffic 542 flows. This issue is further discussed in Section 8. 544 - Soft State: The mechanism must maintain soft state information 545 about the reservations. This means that state information must 546 periodically be refreshed if the reservation is to be maintained; 547 otherwise the state information and corresponding reservations 548 will expire after some pre-specified interval. 550 - Centralized or Distributed Implementation: In the case of a 551 centralized implementation, a single entity manages the resources 552 of the entire subnet. This approach has the advantage of being 553 easier to deploy since bridges and switches may not need to be 554 upgraded with additional functionality. However, this approach 555 scales poorly with geographical size of the subnet and the number 556 of end stations attached. In a fully distributed implementation, 557 each segment will have a local entity managing its resources. 558 This approach has better scalability than the former. However, 559 it requires that all bridges and switches in the network support 560 new mechanisms. It is also possible to have a semi- distributed 561 implementation where there is more than one entity, each managing 562 the resources of a subset of segments and bridges/switches 563 within the subnet. Ideally, implementation should be flexible; 564 i.e. a centralized approach may be used for small subnets and a 565 distributed approach can be used for larger subnets. Examples 566 of centralized and distributed implementations are discussed in 567 Section 6. 569 - Scalability: The mechanism and protocols should have a low 570 overhead and should scale to the largest receiver groups likely 571 to occur within a single link layer domain. 573 - Fault Tolerance and Recovery: The mechanism must be able to 574 function in the presence of failures; i.e. there should not 575 be a single point of failure. For instance, in a centralized 576 implementation, some mechanism must be specified for back-up and 577 recovery in the event of failure. 579 - Interaction with Existing Resource Management Controls: The 580 interaction with existing infrastructure for resource management 581 needs to be specified. For example, FDDI has a resource 582 management mechanism called the "Synchronous Bandwidth Manager". 583 The mechanism must be designed so that it takes advantage of, 584 and specifies the interaction with, existing controls where 585 available. 587 5.2. Goals 589 - Independence from higher layer protocols: The mechanism should, 590 as far as possible, be independent of higher layer protocols such 591 as RSVP and IP. Independence from RSVP is desirable so that it 592 can interwork with other reservation protocols such as ST2 [10]. 593 Independence from IP is desirable so that it can interwork with 594 other network layer protocols such as IPX, NetBIOS, etc. 596 - Receiver heterogeneity: this refers to multicast communication 597 where different receivers request different levels of service. 598 For example, in a multicast group with many receivers, it 599 is possible that one of the receivers desires a lower delay 600 bound than the others. A better delay bound may be provided 601 by increasing the amount of resources reserved along the path 602 to that receiver while leaving the reservations for the other 603 receivers unchanged. In its most complex form, receiver 604 heterogeneity implies the ability to simultaneously provide 605 various levels of service as requested by different receivers. 606 In its simplest form, receiver heterogeneity will allow a 607 scenario where some of the receivers use best effort service and 608 those requiring service guarantees make a reservation. Receiver 609 heterogeneity, especially for the reserved/best effort scenario, 610 is a very desirable function. More details on supporting 611 receiver heterogeneity are provided in Section 8. 613 - Support for different filter styles: It is desirable to provide 614 support for the different filter styles defined by RSVP such as 615 fixed filter, shared explicit and wildcard. Some of the issues 616 with respect to supporting such filter styles in the link layer 617 domain are examined in Section 8. 619 - Path Selection: In source routed LAN technologies such as 620 Token Ring/IEEE 802.5, it may be useful for the mechanism to 621 incorporate the function of path selection. Using an appropriate 622 path selection mechanism may optimize utilization of network 623 resources. 625 5.3. Non-goals 627 This document describes service mappings onto existing IEEE and ANSI 628 defined standard MAC layers and uses standard MAC layer services 629 as in IEEE 802.1 bridging. It does not attempt to make use of or 630 describe the capabilities of other proprietary or standard MAC layer 631 protocols although it should be noted that published work regarding 632 MAC layers suitable for QoS mappings exists. These are outside the 633 scope of the ISSLL working group charter. 635 5.4. Assumptions 637 This framework assumes that typical subnetworks that are concerned 638 about QoS will be "switch rich"; most communication between 639 end stations using integrated services support is expected to 640 pass through at least one switch. The mechanisms and protocols 641 described will be trivially extensible to communicating systems on 642 the same shared medium, but it is important not to allow problem 643 generalization to complicate the targeted practical application which 644 is switch rich LAN topologies. There have also been developments in 645 the area of MAC enhancements to ensure delay deterministic access on 646 network links e.g. IEEE 802.12 [19] and also proprietary schemes. 648 Although we illustrate most examples for this model using RSVP as 649 the upper layer QoS signaling protocol, there are actually no real 650 dependencies on this protocol. RSVP could be replaced by some other 651 dynamic protocol, or the requests could be made by network management 652 or other policy entities. The SBM signaling protocol [14], which is 653 based upon RSVP, is designed to work seamlessly in the architecture 654 described in this memo. 656 There may be a heterogeneous mix of switches with different 657 capabilities, all compliant with IEEE 802.1D [2,3], but implementing 658 varied queueing and forwarding mechanisms ranging from simple systems 659 with two queues per port and static priority scheduling, to more 660 complex systems with multiple queues using WFQ or other algorithms. 662 The problem is decomposed into smaller independent parts which may 663 lead to sub-optimal use of the network resources but we contend that 664 such benefits are often equivalent to very small improvement in 665 network efficiency in a LAN environment. Therefore, it is a goal 666 that the switches in a network operate using a much simpler set of 667 information than the RSVP engine in a router. In particular, it is 668 assumed that such switches do not need to implement per flow queueing 669 and policing (although they may do so). 671 A fundamental assumption of the IntServ model is that flows are 672 isolated from each other throughout their transit across a network. 673 Intermediate queueing nodes are expected shape or police the traffic 674 to ensure conformance to the negotiated traffic flow specification. 675 In the architecture proposed here for mapping to Layer 2, we 676 diverge from that assumption in the interest of simplicity. The 677 policing/shaping functions are assumed to be implemented in end 678 stations. In some LAN environments, it is reasonable to assume that 679 end stations are trusted to adhere to their negotiated contracts at 680 the inputs to the network, and that we can afford to over-allocate 681 resources during admission control to compensate for the inevitable 682 packet jitter/bunching introduced by the switched network itself. 684 This divergence has some implications on the types of receiver 685 heterogeneity that can be supported and the statistical multiplexing 686 gains that may be exploited, especially for Controlled Load flows. 687 This is discussed in Section 8.7 of this document. 689 6. Basic Architecture 691 The functional requirements described in Section 5 will be performed 692 by an entity which we refer to as the Bandwidth Manager (BM). The BM 693 is responsible for providing mechanisms for an application or higher 694 layer protocol to request QoS from the network. For architectural 695 purposes, the BM consists of the following components. 697 6.1. Components 699 6.1.1. Requester Module 701 The Requester Module (RM) resides in every end station in the subnet. 702 One of its functions is to provide an interface between applications 703 or higher layer protocols such as RSVP, ST2, SNMP, etc. and the BM. 704 An application can invoke the various functions of the BM by using 705 the primitives for communication with the RM and providing it with 706 the appropriate parameters. To initiate a reservation, in the link 707 layer domain, the following parameters must be passed to the RM: the 708 service desired (Guaranteed Service or Controlled Load), the traffic 709 descriptors contained in the TSpec, and an RSpec specifying the 710 amount of resources to be reserved [9]. More information on these 711 parameters may be found in the relevant Integrated Services documents 712 [6,7,8,9]. When RSVP is used for signaling at the network layer, 713 this information is available and needs to be extracted from the RSVP 714 PATH and RSVP RESV messages (See [5] for details). In addition to 715 these parameters, the network layer addresses of the end points must 716 be specified. The RM must then translate the network layer addresses 717 to link layer addresses and convert the request into an appropriate 718 format which is understood by other components of the BM responsible 719 admission control. The RM is also responsible for returning the 720 status of requests processed by the BM to the invoking application or 721 higher layer protocol. 723 6.1.2. Bandwidth Allocator 725 The Bandwidth Allocator (BA) is responsible for performing admission 726 control and maintaining state about the allocation of resources 727 in the subnet. An end station can request various services, e.g. 728 bandwidth reservation, modification of an existing reservation, 729 queries about resource availability, etc. These requests are 730 processed by the BA. The communication between the end station and 731 the BA takes place through the RM. The location of the BA will 732 depend largely on the implementation method. In a centralized 733 implementation, the BA may reside on a single station in the 734 subnet. In a distributed implementation, the functions of the BA 735 may be distributed in all the end stations and bridges/switches as 736 necessary. The BA is also responsible for deciding how to label 737 flows, e.g. based on the admission control decision, the BA may 738 indicate to the RM that packets belonging to a particular flow be 739 tagged with some priority value which maps to the appropriate traffic 740 class. 742 6.1.3. Communication Protocols 744 The protocols for communication between the various components of the 745 BM system must be specified. These include the following: 747 - Communication between the higher layer protocols and the RM: 748 The BM must define primitives for the application to initiate 749 reservations, query the BA about available resources, and 750 change or delete reservations, etc. These primitives could be 751 implemented as an API for an application to invoke functions of 752 the BM via the RM. 754 - Communication between the RM and the BA: A signaling mechanism 755 must be defined for the communication between the RM and the BA. 756 This protocol will specify the messages which must be exchanged 757 between the RM and the BA in order to service various requests by 758 the higher layer entity. 760 - Communication between peer BAs: If there is more than one BA in 761 the subnet, a means must be specified for inter-BA communication. 762 Specifically, the BAs must be able to decide among themselves 763 about which BA would be responsible for which segments and 764 bridges or switches. Further, if a request is made for resource 765 reservation along the domain of multiple BAs, the BAs must be 766 able to handle such a scenario correctly. Inter-BA communication 767 will also be responsible for back-up and recovery in the event of 768 failure. 770 6.2. Centralized vs. Distributed Implementations 772 Example scenarios are provided showing the location of the the 773 components of the bandwidth manager in centralized and fully 774 distributed implementations. Note that in either case, the RM must 775 be present in all end stations which desire to make reservations. 776 Essentially, centralized or distributed refers to the implementation 777 of the BA, the component responsible for resource reservation 778 and admission control. In the figures below, "App" refers to 779 the application making use of the BM. It could either be a user 780 application, or a higher layer protocol process such as RSVP. 782 +---------+ 783 .-->| BA |<--. 784 / +---------+ \ 785 / .-->| Layer 2 |<--. \ 786 / / +---------+ \ \ 787 / / \ \ 788 / / \ \ 789 +---------+ / / \ \ +---------+ 790 | App |<----- /-/---------------------------\-\----->| App | 791 +---------+ / / \ \ +---------+ 792 | RM |<----. / \ .--->| RM | 793 +---------+ / +---------+ +---------+ \ +---------+ 794 | Layer 2 |<------>| Layer 2 |<------>| Layer 2 |<------>| Layer 2 | 795 +---------+ +---------+ +---------+ +---------+ 797 RSVP Host/ Intermediate Intermediate RSVP Host/ 798 Router Bridge/Switch Bridge/Switch Router 800 Figure 1: Bandwidth Manager with centralized Bandwidth Allocator 802 Figure 1 shows a centralized implementation where a single BA is 803 responsible for admission control decisions for the entire subnet. 804 Every end station contains a RM. Intermediate bridges and switches 805 in the network need not have any functions of the BM since they will 806 not be actively participating in admission control. The RM at the 807 end station requesting a reservation initiates communication with 808 its BA. For larger subnets, a single BA may not be able to handle 809 the reservations for the entire subnet. In that case it would be 810 necessary to deploy multiple BAs, each managing the resources of a 811 non-overlapping subset of segments. In a centralized implementation, 812 the BA must have some model of the Layer 2 topology of the subnet 813 e.g. link layer spanning tree information, in order to be able to 814 reserve resources on appropriate segments. Without this topology 815 information, the BM would have to reserve resources on all segments 816 for all flows which, in a switched network, would lead to very 817 inefficient utilization of resources. 819 +---------+ +---------+ 820 | App |<-------------------------------------------->| App | 821 +---------+ +---------+ +---------+ +---------+ 822 | RM/BA |<------>| BA |<------>| BA |<------>| RM/BA | 823 +---------+ +---------+ +---------+ +---------+ 824 | Layer 2 |<------>| Layer 2 |<------>| Layer 2 |<------>| Layer 2 | 825 +---------+ +---------+ +---------+ +---------+ 827 RSVP Host/ Intermediate Intermediate RSVP Host/ 828 Router Bridge/Switch Bridge/Switch Router 830 Figure 2: Bandwidth Manager with fully 831 distributed Bandwidth Allocator 833 Figure 2 depicts the scenario of a fully distributed bandwidth 834 manager. In this case, all devices in the subnet have BM 835 functionality. All the end hosts are still required to have a 836 RM. In addition, all stations actively participate in admission 837 control. With this approach, each BA would need only local topology 838 information since it is responsible for the resources on segments 839 that are directly connected to it. This local topology information, 840 such as a list of ports active on the spanning tree and which unicast 841 addresses are reachable from which ports, is readily available in 842 today's switches. Note that in the figures above, the arrows between 843 peer layers are used to indicate logical connectivity. 845 7. Model of the Bandwidth Manager in a Network 847 In this section we describe how the model above fits with the 848 existing IETF Integrated Services model of IP hosts and routers. 849 First, we describe Layer 3 host and router implementations. Next, we 850 describe how the model is applied in Layer 2 switches. Throughout 851 we indicate any differences between centralized and distributed 852 implementations. 854 7.1. End Station Model 856 7.1.1. Layer 3 Client Model 858 We assume the same client model as IntServ and RSVP where we use the 859 term "client" to mean the entity handling QoS in the Layer 3 device 860 at each end of a Layer 2 hop. In this model, the sending client 861 is responsible for local admission control and packet scheduling 862 onto its link in accordance with the negotiated service. As with 863 the IntServ model, this involves per flow scheduling with possible 864 traffic shaping/policing in every such originating source. 866 For now, we assume that the client runs an RSVP process which 867 presents a session establishment interface to applications, signals 868 over the network, programs a scheduler and classifier in the driver, 869 and interfaces to a policy control module. In particular, RSVP also 870 interfaces to a local admission control module which is the focus of 871 this section. 873 The following figure, reproduced from the RSVP specification, depicts 874 the RSVP process in sending hosts. 876 7.1.2. Requests to Layer 2 ISSLL 878 The local admission control entity within a client is responsible for 879 mapping Layer 3 session establishment requests into Layer 2 language. 881 The upper layer entity makes a request, in generalized terms to ISSLL 882 of the form: 884 "May I reserve for traffic with 885 with from to and 886 how should I label it?" 888 where 889 +-----------------------------+ 890 | +-------+ +-------+ | RSVP 891 | |Appli- | | RSVP <-------------------> 892 | | cation<--> | | 893 | | | |process| +-----+| 894 | +-+-----+ | +->Polcy|| 895 | | +--+--+-+ |Cntrl|| 896 | |data | | +-----+| 897 |===|===========|==|==========| 898 | | +--------+ | +-----+| 899 | | | | +--->Admis|| 900 | +-V--V-+ +---V----+ |Cntrl|| 901 | |Class-| | Packet | +-----+| 902 | | ifier|==>Schedulr|===================> 903 | +------+ +--------+ | data 904 +-----------------------------+ 906 Figure 3: RSVP in Sending Hosts 908 = Sender Tspec (e.g. bandwidth, burstiness, 909 MTU) 910 = FlowSpec (e.g. latency, jitter bounds) 911 = IP address(es) 912 = IP address(es) - may be multicast 914 7.1.3. At the Layer 3 Sender 916 The ISSLL functionality in the sender is illustrated in Figure 4. 918 The functions of the Requester Module may be summarized as follows: 920 - Maps the endpoints of the conversation to Layer 2 addresses 921 in the LAN, so that the client can determine what traffic is 922 going where. This function probably makes reference to the ARP 923 protocol cache for unicast or performs an algorithmic mapping for 924 multicast destinations. 926 - Communicates with any local Bandwidth Allocator module for local 927 admission control decisions. 929 - Formats a SBM request to the network with the mapped addresses 930 and flow/filter specs. 932 from IP from RSVP 933 +----|------------|------------+ 934 | +--V----+ +---V---+ | 935 | | Addr <---> | | SBM signaling 936 | |mapping| |Request|<-----------------------> 937 | +---+---+ |Module | | 938 | | | | | 939 | +---+---+ | | | 940 | | 802 <---> | | 941 | | header| +-+-+-+-+ | 942 | +--+----+ / | | | 943 | | / | | +-----+ | 944 | | +-----+ | +->|Band-| | 945 | | | | |width| | 946 | +--V-V-+ +-----V--+ |Alloc| | 947 | |Class-| | Packet | +-----+ | 948 | | ifier|==>Schedulr|=========================> 949 | +------+ +--------+ | data 950 +------------------------------+ 952 Figure 4: ISSLL in a Sending End Station 954 - Receives a response from the network and reports the admission 955 control decision to the higher layer entity, along with any 956 negotiated modifications to the session parameters. 958 - Saves any returned user_priority to be associated with this 959 session in a "802 header" table. This will be used when 960 constructing the Layer 2 headers for future data packets 961 belonging to this session. This table might, for example, be 962 indexed by the RSVP flow identifier. 964 The Bandwidth Allocator (BA) component is only present when a 965 distributed BA model is implemented. When present, its function is 966 basically to apply local admission control for the outgoing link 967 bandwidth and driver's queueing resources. 969 7.1.4. At the Layer 3 Receiver 971 The ISSLL functionality in the receiver is simpler is illustrated in 972 Figure 5. 974 The functions of the Requester Module may be summarized as follows: 976 to RSVP to IP 977 ^ ^ 978 +----|------------|------+ 979 | +--+----+ | | 980 SBM signaling | |Request| +---+---+ | 981 <-------------> |Module | | Strip | | 982 | +--+---++ |802 hdr| | 983 | | \ +---^---+ | 984 | +--v----+\ | | 985 | | Band- | \ | | 986 | | width| \ | | 987 | | Alloc | . | | 988 | +-------+ | | | 989 | +------+ +v---+----+ | 990 data | |Class-| | Packet | | 991 <==============>| ifier|==>|Scheduler| | 992 | +------+ +---------+ | 993 +------------------------+ 995 Figure 5: ISSLL in a Receiving End Station 997 - Handles any received SBM protocol indications. 999 - Communicates with any local BA for local admission control 1000 decisions. 1002 - Passes indications up to RSVP if OK. 1004 - Accepts confirmations from RSVP and relays them back via SBM 1005 signaling towards the requester. 1007 - May program a receive classifier and scheduler, if used, to 1008 identify traffic classes of received packets and accord them 1009 appropriate treatment e.g. reservation of buffers for particular 1010 traffic classes. 1012 - Programs the receiver to strip away link layer header information 1013 from received packets. 1015 The Bandwidth Allocator, present only in a distributed implementation 1016 applies local admission control to see if a request can be supported 1017 with appropriate local receive resources. 1019 7.2. Switch Model 1021 7.2.1. Centralized Bandwidth Allocator 1023 Where a centralized Bandwidth Allocator model is implemented, 1024 switches do not take part in the admission control process. 1025 Admission control is implemented by a centralized BA, e.g. a "Subnet 1026 Bandwidth Manager" (SBM) as described in [14]. This centralized BA 1027 may actually be co-located with a switch but its functions would 1028 not necessarily then be closely tied with the switch's forwarding 1029 functions as is the case with the distributed BA described below. 1031 7.2.2. Distributed Bandwidth Allocator 1033 The model of Layer 2 switch behavior described here uses the 1034 terminology of the SBM protocol as an example of an admission control 1035 protocol. The model is equally applicable when other mechanisms, 1036 e.g. static configuration or network management, are in use for 1037 admission control. We define the following entities within the 1038 switch: 1040 - Local Admission Control Module: One of these on each port 1041 accounts for the available bandwidth on the link attached to that 1042 port. For half duplex links, this involves taking account of the 1043 resources allocated to both transmit and receive flows. For full 1044 duplex links, the input port accountant's task is trivial. 1046 - Input SBM Module: One instance on each port performs the 1047 "network" side of the signaling protocol for peering with clients 1048 or other switches. It also holds knowledge about the mappings of 1049 IntServ classes to user_priority. 1051 - SBM Propagation Module: Relays requests that have passed 1052 admission control at the input port to the relevant output ports' 1053 SBM modules. This will require access to the switch's forwarding 1054 table (Layer-2 "routing table" cf. RSVP model) and port spanning 1055 tree state. 1057 - Output SBM Module: Forwards requests to the next Layer 2 or 1058 Layer 3 hop. 1060 - Classifier, Queue and Scheduler Module: The functions of this 1061 module are basically as described by the Forwarding Process of 1062 IEEE 802.1D (see Section 3.7 of [3]). The Classifier module 1063 identifies the relevant QoS information from incoming packets and 1064 uses this, together with the normal bridge forwarding database, 1065 to decide at which output port and traffic class to enqueue 1066 the packet. Different types of switches will use different 1067 techniques for flow identification (see Section 8.1) In IEEE 1068 802.1D switches this information is the regenerated user_priority 1069 parameter which has already been decoded by the receiving MAC 1070 service and potentially remapped by the forwarding process (see 1071 section 3.7.3 of [3]). This does not preclude more sophisticated 1072 classification rules such as the classification of individual 1073 IntServ flows. The Queue and Scheduler hold the output queues 1074 for ports and provide the algorithm for servicing the queues 1075 for transmission onto the output link in order to provide the 1076 promised IntServ service. Switches will implement one or more 1077 output queues per port and all will implement at least a basic 1078 static priority dequeueing algorithm as their default, in 1079 accordance with IEEE 802.1D. 1081 - Ingress Traffic Class Mapping and Policing Module: Its functions 1082 are as described in IEEE 802.1D Section 3.7. This optional 1083 module may police the data within traffic classes for conformance 1084 to the negotiated parameters, and may discard packets or re-map 1085 the user_priority. The default behavior is to pass things 1086 through unchanged. 1088 - Egress Traffic Class Mapping Module: Its functions are as 1089 described in IEEE 802.1D Section 3.7. This optional module may 1090 perform re-mapping of traffic classes on a per output port basis. 1091 The default behavior is to pass things through unchanged. 1093 Figure 6 shows all of the modules in an ISSLL enabled switch. The 1094 ISSLL model is a superset of the IEEE 802.1D bridge model. 1096 7.3. Admission Control 1098 On receipt of an admission control request, a switch performs the 1099 following actions, again using SBM as an example. The behavior 1100 is different depending on whether the "Designated SBM" for this 1101 segment is within this switch or not. See [14] for a more detailed 1102 specification of the DSBM/SBM actions. 1104 - If the ingress SBM is the "Designated SBM" for this link, it 1105 either translates any received user_priority or selects a Layer 1106 2 traffic class which appears compatible with the request and 1107 whose use does not violate any administrative policies in force. 1108 In effect, it matches the requested service with the available 1109 traffic classes and chooses the "best" one. It ensures that, 1110 if this reservation is successful, the value of user_priority 1111 corresponding to that traffic class is passed back to the client. 1113 +-------------------------------+ 1114 SBM signaling | +-----+ +------+ +------+ | SBM signaling 1115 <------------------>| IN |<->| SBM |<->| OUT |<----------------> 1116 | | SBM | | prop.| | SBM | | 1117 | +-++--+ +---^--+ /----+-+ | 1118 | / | | / | | 1119 ______________| / | | | | +-------------+ 1120 | \ /+--V--+ | | +--V--+ / | 1121 | \ ____/ |Local| | | |Local| / | 1122 | \ / |Admis| | | |Admis| / | 1123 | \/ |Cntrl| | | |Cntrl| / | 1124 | +-----V+\ +-----+ | | +-----+ /+-----+ | 1125 | |traff | \ +---+--+ +V-------+ / |egrss| | 1126 | |class | \ |Filter| |Queue & | / |traff| | 1127 | |map & |=====|==========>|Data- |=| Packet |=|===>|class| | 1128 | |police| | | base| |Schedule| | |map | | 1129 | +------+ | +------+ +--------+ | +-+---+ | 1130 +----^---------+-------------------------------+------|------+ 1131 data in | |data out 1132 ========+ +========> 1134 Figure 6: ISSLL in a Switch 1136 - The ingress DSBM observes the current state of allocation of 1137 resources on the input port/link and then determines whether 1138 the new resource allocation from the mapped traffic class can 1139 be accommodated. The request is passed to the reservation 1140 propagator if accepted. 1142 - If the ingress SBM is not the "Designated SBM" for this link then 1143 it directly passes the request on to the reservation propagator. 1145 - The reservation propagator relays the request to the bandwidth 1146 accountants on each of the switch's outbound links to which 1147 this reservation would apply. This implies an interface to 1148 routing/forwarding database. 1150 - The egress bandwidth accountant observes the current state 1151 of allocation of queueing resources on its outbound port and 1152 bandwidth on the link itself and determines whether the new 1153 allocation can be accommodated. Note that this is only a local 1154 decision at this switch hop; further Layer 2 hops through the 1155 network may veto the request as it passes along. 1157 - The request, if accepted by this switch, is propagated on 1158 each output link selected. Any user_priority described in the 1159 forwarded request must be translated according to any egress 1160 mapping table. 1162 - If accepted, the switch must notify the client of the 1163 user_priority to be used for packets belonging to that flow. 1164 Again, this is an optimistic approach assuming that admission 1165 control succeeds; downstream switches may refuse the request. 1167 - If this switch wishes to reject the request, it can do so by 1168 notifying the original client by means of its Layer 2 address. 1170 7.4. QoS Signaling 1172 The mechanisms described in this document make use of a signaling 1173 protocol for devices to communicate their admission control requests 1174 across the network. The service definitions to be provided by 1175 such a protocol e.g. [14] are described below. We illustrate the 1176 primitives and information that need to be exchanged with such a 1177 signaling protocol entity. In all of the examples, appropriate 1178 delete/cleanup mechanisms will also have to be provided for tearing 1179 down established sessions. 1181 7.4.1. Client Service Definitions 1183 The following interfaces can be identified from Figures 4 and 5. 1185 - SBM <-> Address Mapping 1187 This is a simple lookup function which may require ARP protocol 1188 interactions or an algorithmic mapping. The Layer 2 addresses 1189 are needed by SBM for inclusion in its signaling messages to 1190 avoid requiring that switches participating in the signaling have 1191 Layer 3 information to perform the mapping. 1193 l2_addr = map_address( ip_addr ) 1195 - SBM <-> Session/Link Layer Header 1197 This is for notifying the transmit path of how to add Layer 2 1198 header information, e.g. user_priority values to the traffic 1199 of each outgoing flow. The transmit path will provide the 1200 user_priority value when it requests a MAC layer transmit 1201 operation for each packet. The user_priority is one of the 1202 parameters passed in the packet transmit primitive defined by the 1203 IEEE 802 service model. 1205 bind_l2_header( flow_id, user_priority ) 1207 - SBM <-> Classifier/Scheduler 1209 This is for notifying transmit classifier/scheduler of any 1210 additional Layer 2 information associated with scheduling the 1211 transmission of a packet flow. This primitive may be unused in 1212 some implementations or it may be used, for example, to provide 1213 information to a transmit scheduler that is performing per 1214 traffic class scheduling in addition to the per flow scheduling 1215 required by IntServ; the Layer 2 header may be a pattern (in 1216 addition to the FilterSpec) to be used to identify the flow's 1217 traffic. 1219 bind_l2schedulerinfo( flow_id, , l2_header, traffic_class ) 1221 - SBM <-> Local Admission Control 1223 This is used for applying local admission control for a session 1224 e.g. is there enough transmit bandwidth still uncommitted 1225 for this potential new session? Are there sufficient receive 1226 buffers? This should commit the necessary resources if it 1227 succeeds. It will be necessary to release these resources at 1228 a later stage if the admission control fails at a later stage. 1229 This call would be made, for example, by a segment's Designated 1230 SBM. 1232 status = admit_l2session( flow_id, Tspec, FlowSpec ) 1234 - SBM <-> RSVP 1236 This is outlined above in Section 7.1.2 and fully described in 1237 [14]. 1239 - Management Interfaces 1241 Some or all of the modules described by this model will also 1242 require configuration management. It is expected that details of 1243 the manageable objects will be specified by future work in the 1244 ISSLL WG. 1246 7.4.2. Switch Service Definitions 1248 The following interfaces are identified from Figure 6. 1250 - SBM <-> Classifier 1252 This is for notifying the receive classifier of how to match 1253 incoming Layer 2 information with the associated traffic class. 1254 It may in some cases consist of a set of read only default 1255 mappings. 1257 bind_l2classifierinfo( flow_id, l2_header, traffic_class ) 1259 - SBM <-> Queue and Packet Scheduler 1261 This is for notifying transmit scheduler of additional Layer 2 1262 information associated with a given traffic class. It may be 1263 unused in some cases (see discussion in previous section). 1265 bind_l2schedulerinfo( flow_id, l2_header, traffic_class ) 1267 - SBM <-> Local Admission Control 1269 Same as for the host discussed above. 1271 - SBM <-> Traffic Class Map and Police 1273 Optional configuration of any user_priority remapping that 1274 might be implemented on ingress to and egress from the ports of 1275 a switch. For IEEE 802.1D switches, it is likely that these 1276 mappings will have to be consistent across all ports. 1278 bind_l2ingressprimap( inport, in_user_pri, internal_priority ) 1279 bind_l2egressprimap( outport, internal_priority, out_user_pri ) 1281 Optional configuration of any Layer 2 policing function to be 1282 applied on a per class basis to traffic matching the Layer 2 1283 header. If the switch is capable of per flow policing then 1284 existing IntServ/RSVP models will provide a service definition 1285 for that configuration. 1287 bind_l2policing( flow_id, l2_header, Tspec, FlowSpec ) 1289 - SBM <-> Filtering Database 1291 SBM propagation rules need access to the Layer 2 forwarding 1292 database to determine where to forward SBM messages. This is 1293 analogous to RSRR interface in Layer 3 RSVP. 1295 output_portlist = lookup_l2dest( l2_addr ) 1297 - Management Interfaces 1298 Some or all of the modules described by this model will also 1299 require configuration management. It is expected that details of 1300 the manageable objects will be specified by future work in the 1301 ISSLL working group. 1303 8. Implementation Issues 1305 As stated earlier, the Integrated Services working group has defined 1306 various service classes offering varying degrees of QoS guarantees. 1307 Initial effort will concentrate on enabling the Controlled Load [6] 1308 and Guaranteed Service classes [7]. The Controlled Load service 1309 provides a loose guarantee, informally stated as "the same as best 1310 effort would be on an unloaded network". The Guaranteed Service 1311 provides an upper bound on the transit delay of any packet. The 1312 extent to which these services can be supported at the link layer 1313 will depend on many factors including the topology and technology 1314 used. Some of the mapping issues are discussed below in light of 1315 the emerging link layer standards and the functions supported by 1316 higher layer protocols. Considering the limitations of some of the 1317 topologies, it may not be possible to satisfy all the requirements 1318 for Integrated Services on a given topology. In such cases, it 1319 is useful to consider providing support for an approximation of 1320 the service which may suffice in most practical instances. For 1321 example, it may not be feasible to provide policing/shaping at each 1322 network element (bridge/switch) as required by the Controlled Load 1323 specification. But if this task is left to the end stations, a 1324 reasonably good approximation to the service can be obtained. 1326 8.1. Switch Characteristics 1328 There are many LAN bridges/switches with varied capabilities for 1329 supporting QoS. We discuss below the various kinds of devices that 1330 that one may expect to find in a LAN environment. 1332 The most basic bridge is one which conforms to IEEE 802.1D [2]. This 1333 device has a single queue per output port, and uses the spanning tree 1334 algorithm to eliminate topology loops. Networks constructed from 1335 this kind of device cannot be expected to provide service guarantees 1336 of any kind because of the complete lack of traffic isolation. 1338 The next level of bridges/switches are those which conform to the 1339 more recently revised IEEE 802.1D specification. It will include 1340 support for queueing up to eight traffic classes separately. The 1341 level of traffic isolation provided is coarse because all flows 1342 corresponding to a particular traffic class are aggregated. Further, 1343 it is likely that more than one priority will map to a traffic class 1344 depending on the number of queues implemented in the switch. It 1345 would be difficult for such a device to offer protection against 1346 misbehaving flows. The scope of multicast traffic may be limited by 1347 using GMRP to only those segments which are on the path to interested 1348 receivers. 1350 A next step above these devices are bridges/switches which implement 1351 optional parts of the IEEE 802.1D specification such as mapping the 1352 received user_priority to some internal set of canonical values 1353 on a per-input-port basis. It may also support the mapping of 1354 these internal canonical values onto transmitted user_priority on 1355 a per-output-port basis. With these extra capabilities, network 1356 administrators can perform mapping of traffic classes between 1357 specific pairs of ports, and in doing so gains more control over 1358 admission to traffic into the protected classes. 1360 Other entirely optional features that some bridges/switches may 1361 support include classification of IntServ flows using fields in the 1362 network layer header, per-flow policing and/or reshaping which is 1363 essential for supporting Guaranteed Service, and more sophisticated 1364 scheduling algorithms such as variants of weighted fair queueing to 1365 limit the bandwidth consumed by a traffic class. Note that it is 1366 advantageous to perform flow isolation and for all network elements 1367 to police each flow in order to support the Controlled Load and 1368 Guaranteed Service. 1370 8.2. Queueing 1372 Connectionless packet networks in general, and LANs in particular, 1373 work today because of scaling choices in network provisioning. 1374 Typically, excess bandwidth and buffering is provisioned in the 1375 network to absorb the traffic sourced by higher layer protocols, 1376 often sufficient to cause their transmission windows to run out on a 1377 statistical basis, so that network overloads are rare and transient 1378 and the expected loading is very low. 1380 With the advent of time-critical traffic such over-provisioning 1381 has become far less easy to achieve. Time-critical frames may be 1382 queued for annoyingly long periods of time behind temporary bursts 1383 of file transfer traffic, particularly at network bottleneck points, 1384 e.g. at the 100 Mbps to 10 Mbps transition that might occur between 1385 the riser to the wiring closet and the final link to the user from 1386 a desktop switch. In this case, however, if it is known a priori 1387 (either by application design, on the basis of statistics, or on 1388 by administrative control), that time-critical traffic is a small 1389 fraction of the total bandwidth, it suffices to give it strict 1390 priority over the non-time-critical traffic. The worst case delay 1391 experienced by the time-critical traffic is roughly the maximum 1392 transmission time of a maximum length non-time-critical frame -- less 1393 than a millisecond for 10 Mbps Ethernet, and well below the end to 1394 end delay budget based on human perception times. 1396 When more than one priority service is to be offered by a network 1397 element e.g. one which supports both Controlled Load as well as 1398 Guaranteed Service, the requirements for the scheduling discipline 1399 becomes more complex. In order to provide the required isolation 1400 between the service classes, it will probably be necessary to queue 1401 them separately. There is then an issue of how to service the 1402 queues which requires a combination of admission control and more 1403 intelligent queueing disciplines. As with the service specifications 1404 themselves, the specification of queueing algorithms is beyond the 1405 scope of this document. 1407 8.3. Mapping of Services to Link Level Priority 1409 The number of traffic classes supported and access methods of the 1410 technology under consideration will determine how many and what 1411 services may be supported. Native Token Ring/IEEE 802.5, for 1412 instance, supports eight priority levels which may be mapped to 1413 one or more traffic classes. Ethernet/IEEE 802.3 has no support 1414 for signaling priorities within frames. However, the IEEE 802 1415 standards committee has recently developed a new standard for 1416 bridges/switches related to multimedia traffic expediting and 1417 dynamic multicast filtering [3]. A packet format for carrying a 1418 user_priority field on all IEEE 802 LAN media types is now defined 1419 in [4]. These standards allow for up to eight traffic classes 1420 on all media. The user_priority bits carried in the frame are 1421 mapped to a particular traffic class within a bridge/switch. The 1422 user_priority is signaled on an end-to-end basis, unless overridden 1423 by bridge/switch management. The traffic class that is used by a 1424 flow should depend on the quality of service desired and whether the 1425 reservation is successful or not. Therefore, a sender should use the 1426 user_priority value which maps to the best effort traffic class until 1427 told otherwise by the BM. The BM will, upon successful completion of 1428 resource reservation, specify the value of user_priority to be used 1429 by the sender for that session's data. An accompanying memo [13] 1430 addresses the issue of mapping the various Integrated Services to 1431 appropriate traffic classes. 1433 8.4. Re-mapping of Non-conforming Aggregated Flows 1435 One other topic under discussion in the IntServ context is how to 1436 handle the traffic for data flows from sources that exceed their 1437 negotiated traffic contract with the network. An approach that shows 1438 some promise is to treat such traffic with "somewhat less than best 1439 effort" service in order to protect traffic that is normally given 1440 "best effort" service from having to back off. Best effort traffic 1441 is often adaptive, using TCP or other congestion control algorithms, 1442 and it would be unfair to penalize those flows due to badly behaved 1443 traffic from reserved flows which are often set up by non-adaptive 1444 applications. 1446 A possible solution might be to assign normal best effort traffic 1447 to one user_priority and to label excess non-conforming traffic 1448 as a lower user_priority although the re-ordering problems that 1449 might arise from doing this may make this solution undesirable, 1450 particularly if the flows are using TCP. For this reason the 1451 controlled load service recommends dropping excess traffic, rather 1452 than re-mapping to a lower priority. This is further discussed 1453 below. 1455 8.5. Override of Incoming User Priority 1457 In some cases, a network administrator may not trust the 1458 user_priority values contained in packets from a source and may wish 1459 to map these into some more suitable set of values. Alternatively, 1460 due perhaps to equipment limitations or transition periods, values 1461 may need to be re-mapped as the data flows to/from different regions 1462 of a network. 1464 Some switches may implement such a function on input that maps 1465 received user_priority to some internal set of values. This 1466 function is provided by a table known in IEEE 802.1D as the User 1467 Priority Regeneration Table (Table 3-1 in [3]). These values can 1468 then be mapped using an output table described above onto outgoing 1469 user_priority values. These same mappings must also be used when 1470 applying admission control to requests that use the user_priority 1471 values (see e.g. [14]). More sophisticated approaches are also 1472 possible where a device polices traffic flows and adjusts their 1473 onward user_priority based on their conformance to the admitted 1474 traffic flow specifications. 1476 8.6. Different Reservation Styles 1478 In the figure above, SW is a bridge/switch in the link layer domain. 1479 S1, S2, S3, R1 and R2 are end stations which are members of a group 1480 associated with the same RSVP flow. S1, S2 and S3 are upstream 1481 end stations. R1 and R2 are the downstream end stations which 1482 receive traffic from all the senders. RSVP allows receivers R1 and 1483 +-----+ +-----+ +-----+ 1484 | S1 | | S2 | | S3 | 1485 +-----+ +-----+ +-----+ 1486 | | | 1487 | v | 1488 | +-----+ | 1489 +--------->| SW |<---------+ 1490 +-----+ 1491 | | 1492 +----+ +----+ 1493 | | 1494 v V 1495 +-----+ +-----+ 1496 | R1 | | R2 | 1497 +-----+ +-----+ 1499 Figure 7: Illustration of filter styles 1501 R2 to specify reservations which can apply to: (a) one specific 1502 sender only (fixed filter); (b) any of two or more explicitly 1503 specified senders (shared explicit filter); and (c) any sender in 1504 the group (shared wildcard filter). Support for the fixed filter 1505 style is straightforward; a separate reservation is made for the 1506 traffic from each of the senders. However, support for the other 1507 two filter styles has implications regarding policing; i.e. the 1508 merged flow from the different senders must be policed so that they 1509 conform to traffic parameters specified in the filter's RSpec. This 1510 scenario is further complicated if the services requested by R1 and 1511 R2 are different. Therefore, in the absence of policing within 1512 bridges/switches, it may be possible to support only fixed filter 1513 reservations at the link layer. 1515 8.7. Receiver Heterogeneity 1517 At Layer 3, the IntServ model allows heterogeneous receivers 1518 for multicast flows where different branches of a tree can have 1519 different types of reservations for a given multicast destination. 1520 It also supports the notion that trees may have some branches with 1521 reserved flows and some using best effort service. If we were 1522 to treat a Layer 2 subnet as a single network element as defined 1523 in [8], then all of the branches of the distribution tree that 1524 lie within the subnet could be assumed to require the same QoS 1525 treatment and be treated as an atomic unit as regards admission 1526 control, etc. With this assumption, the model and protocols already 1527 defined by IntServ and RSVP already provide sufficient support for 1528 multicast heterogeneity. Note, however, that an admission control 1529 request may well be rejected because just one link in the subnet is 1530 oversubscribed leading to rejection of the reservation request for 1531 the entire subnet. 1533 +-----+ 1534 | S | 1535 +-----+ 1536 | 1537 v 1538 +-----+ +-----+ +-----+ 1539 | R1 |<-----| SW |----->| R2 | 1540 +-----+ +-----+ +-----+ 1542 Figure 8: Example of receiver heterogeneity 1544 As an example, consider Figure 8, SW is a Layer 2 device 1545 (bridge/switch) participating in resource reservation, S is the 1546 upstream source end station and R1 and R2 are downstream end station 1547 receivers. R1 would like to make a reservation for the flow while R2 1548 would like to receive the flow using best effort service. S sends 1549 RSVP PATH messages which are multicast to both R1 and R2. R1 sends 1550 an RSVP RESV message to S requesting the reservation of resources. 1552 If the reservation is successful at Layer 2, the frames addressed to 1553 the group will be categorized in the traffic class corresponding to 1554 the service requested by R1. At SW, there must be some mechanism 1555 which forwards the packet providing service corresponding to the 1556 reserved traffic class at the interface to R1 while using the best 1557 effort traffic class at the interface to R2. This may involve 1558 changing the contents of the frame itself, or ignoring the frame 1559 priority at the interface to R2. 1561 Another possibility for supporting heterogeneous receivers would 1562 be to have separate groups with distinct MAC addresses, one for 1563 each class of service. By default, a receiver would join the "best 1564 effort" group where the flow is classified as best effort. If the 1565 receiver makes a reservation successfully, it can be transferred to 1566 the group for the class of service desired. The dynamic multicast 1567 filtering capabilities of bridges and switches implementing the IEEE 1568 802.1D standard would be a very useful feature in such a scenario. 1570 A given flow would be transmitted only on those segments which are 1571 on the path between the sender and the receivers of that flow. The 1572 obvious disadvantage of such an approach is that the sender needs to 1573 send out multiple copies of the same packet corresponding to each 1574 class of service desired thus potentially duplicating the traffic on 1575 a portion of the distribution tree. 1577 The above approaches would provide very sub-optimal utilization of 1578 resources given the expected size and complexity of the Layer 2 1579 subnets. Therefore, it is desirable to enable switches to apply QoS 1580 differently on different egress branches of a tree that divide at 1581 that switch. 1583 IEEE 802.1D specifies a basic model for multicast whereby a switch 1584 makes multicast forwarding decisions based on the destination 1585 address. This would produce a list of output ports to which the 1586 packet should be forwarded. In its default mode, such a switch 1587 would use the user_priority value in received packets, or a value 1588 regenerated on a per input port basis in the absence of an explicit 1589 value, to enqueue the packets at each output port. Any IEEE 802.1D 1590 switch which supports multiple traffic classes can support this 1591 operation. 1593 If a switch selects per port output queues based only on the incoming 1594 user_priority, as described by IEEE 802.1D, it must treat all 1595 branches of all multicast sessions within that user_priority class 1596 with the same queueing mechanism. Receiver heterogeneity is then 1597 not possible and this could well lead to the failure of an admission 1598 control request for the whole multicast session due to a single 1599 link being oversubscribed Note that in the Layer 2 case as distinct 1600 from the Layer 3 case with RSVP/IntServ, the option of having some 1601 receivers getting the session with the requested QoS and some getting 1602 it best effort does not exist as basic IEEE 802.1 switches are unable 1603 to re-map the user_priority on a per link basis. This could become 1604 an issue with heavy use of dynamic multicast sessions. If a switch 1605 were to implement a separate user_priority mapping at each output 1606 port, then, in some cases, reservations can use a different traffic 1607 class on different paths that branch at such a switch in order to 1608 provide multiple receivers with different QoS. This is possible if 1609 all flows within a traffic class at the ingress to a switch egress 1610 in the same traffic class on a port. For example, traffic may be 1611 forwarded using user_priority 4 on one branch where receivers have 1612 performed admission control and as user_priority 0 on ones where 1613 they have not. We assume that per user_priority queueing without 1614 taking account of input or output ports is the minimum standard 1615 functionality for switches in a LAN environment (IEEE 802.1D) 1616 but that more functional Layer 2 or even Layer 3 switches (i.e. 1617 routers) can be used if even more flexible forms of heterogeneity are 1618 considered necessary to achieve more efficient resource utilization. 1619 The behavior of Layer 3 switches in this context is already well 1620 standardized by the IETF. 1622 9. Network Topology Scenarios 1624 The extent to which service guarantees can be provided by a 1625 network depend to a large degree on the ability to provide the key 1626 functions of flow identification and scheduling in addition to 1627 admission control and policing. This section discusses some of the 1628 capabilities of the LAN technologies under consideration and provides 1629 a taxonomy of possible topologies emphasizing the capabilities 1630 of each with regard to supporting the above functions. For the 1631 technologies considered here, the basic topology of a LAN may be 1632 shared, switched half duplex or switched full duplex. In the shared 1633 topology, multiple senders share a single segment. Contention for 1634 media access is resolved using protocols such as CSMA/CD in Ethernet 1635 and token passing in Token Ring and FDDI. Switched half duplex, 1636 is essentially a shared topology with the restriction that there 1637 are only two transmitters contending for resources on any segment. 1638 Finally, in a switched full duplex topology, a full bandwidth path is 1639 available to the transmitter at each end of the link at all times. 1640 Therefore, in this topology, there is no need for any access control 1641 mechanism such as CSMA/CD or token passing as there is no contention 1642 between the transmitters. Obviously, this topology provides the best 1643 QoS capabilities. Another important element in the discussion of 1644 topologies is the presence or absence of support for multiple traffic 1645 classes. These were discussed earlier in Section 4.1. Depending on 1646 the basic topology used and the ability to support traffic classes, 1647 we identify six scenarios as follows: 1649 1. Shared topology without traffic classes. 1650 2. Shared topology with traffic classes. 1651 3. Switched half duplex topology without traffic classes. 1652 4. Switched half duplex topology with traffic classes. 1653 5. Switched full duplex topology without traffic classes. 1654 6. Switched full duplex topology with traffic classes. 1656 There is also the possibility of hybrid topologies where two or more 1657 of the above coexist. For instance, it is possible that within a 1658 single subnet, there are some switches which support traffic classes 1659 and some which do not. If the flow in question traverses both 1660 kinds of switches in the network, the least common denominator will 1661 prevail. In other words, as far as that flow is concerned, the 1662 network is of the type corresponding to the least capable topology 1663 that is traversed. In the following sections, we present these 1664 scenarios in further detail for some of the different IEEE 802 1665 network types with discussion of their abilities to support the 1666 IntServ services. 1668 9.1. Full Duplex Switched Networks 1670 On a full duplex switched LAN, the MAC protocol is unimportant 1671 as far as access is concerned, but must be factored in to the 1672 characterization parameters advertised by the device since the 1673 access latency is equal to the time required to transmit the largest 1674 packet. Approximate values for the characteristics on various media 1675 are provided in the following tables. These delays should be also 1677 Table 4: Full duplex switched media access latency 1679 -------------------------------------------------- 1680 Type Speed Max Pkt Max Access 1681 Length Latency 1682 -------------------------------------------------- 1683 Ethernet 10 Mbps 1.2 ms 1.2 ms 1684 100 Mbps 120 us 120 us 1685 1 Gbps 12 us 12 us 1686 Token Ring 4 Mbps 9 ms 9 ms 1687 16 Mbps 9 ms 9 ms 1688 FDDI 100 Mbps 360 us 8.4 ms 1689 Demand Priority 100 Mbps 120 us 120 us 1690 -------------------------------------------------- 1692 be considered in the context of the speed of light delay which is 1693 approximately 400 ns for typical 100 m UTP links and 7 us for typical 1694 2 km multimode fiber links. 1696 Full duplex switched network topologies offer good QoS capabilities 1697 for both Controlled Load and Guaranteed Service when supported by 1698 suitable queueing strategies in the switches. 1700 9.2. Shared Media Ethernet Networks 1702 Thus far, we have not discussed the difficulty of dealing with 1703 allocation on a single shared CSMA/CD segment. As soon as any 1704 CSMA/CD algorithm is introduced the ability to provide any form of 1705 Guaranteed Service is seriously compromised in the absence of any 1706 tight coupling between the multiple senders on the link. There are a 1707 number of reasons for not offering a better solution to this problem. 1709 Firstly, we do not believe this is a truly solvable problem as 1710 it would require changes to the MAC protocol. IEEE 802.1 has 1711 examined research showing disappointing simulation results for 1712 performance guarantees on shared CSMA/CD Ethernet without MAC 1713 enhancements. There have been proposals for enhancements to the 1714 MAC layer protocols, e.g. BLAM and enhanced flow control in IEEE 1715 802.3. However, any solution involving an enhanced software MAC 1716 running above the traditional IEEE 802.3 MAC, or other proprietary 1717 MAC protocols, is outside the scope of the ISSLL working group and 1718 this document. Secondly, we are not convinced that it is really an 1719 interesting problem. While there will be end stations on repeated 1720 segments for some time to come, the number of deployed switches is 1721 steadily increasing relative to the number of stations on repeated 1722 segments. This trend is proceeding to the point where it may be 1723 satisfactory to have a solution which assumes that any network 1724 communication requiring resource reservations will take place 1725 through at least one switch or router. Put another way, the easiest 1726 upgrade to existing Layer 2 infrastructure for QoS support is the 1727 installation of segment switching. Only when this has been done 1728 is it worthwhile to investigate more complex solutions involving 1729 admission control. Thirdly, there core of campus networks typically 1730 consists of solutions based on switches rather than on repeated 1731 segments. There may be special circumstances in the future, e.g. 1732 Gigabit buffered repeaters, but the characteristics of these devices 1733 are different from existing CSMA/CD repeaters anyway. 1735 Table 5: Shared Ethernet media access latency 1737 -------------------------------------------------- 1738 Type Speed Max Pkt Max Access 1739 Length Latency 1740 -------------------------------------------------- 1741 Ethernet 10 Mbps 1.2 ms unbounded 1742 100 Mbps 120 us unbounded 1743 1 Gbps 12 us unbounded 1744 -------------------------------------------------- 1746 9.3. Half Duplex Switched Ethernet Networks 1748 Many of the same arguments for sub optimal support of Guaranteed 1749 Service on shared media Ethernet also apply to half duplex switched 1750 Ethernet. In essence, this topology is a medium that is shared 1751 between at least two senders contending for packet transmission. 1752 Unless these are tightly coupled and cooperative, there is always the 1753 chance that the best effort traffic of one will interfere with the 1754 reserved traffic of the other. Dealing with such a coupling would 1755 seem to require some form of modification to the MAC protocol. 1757 Not withstanding the above argument, half duplex switched topologies 1758 do seem to offer the chance to provide Controlled Load service. With 1759 the knowledge that there are exactly two potential senders that are 1760 both using prioritization for their Controlled Load traffic over best 1761 effort flows, and with admission control having been done for those 1762 flows based on that knowledge, the media access characteristics while 1763 not deterministic are somewhat predictable. This is probably a close 1764 enough useful approximation to the Controlled Load service. 1766 Table 6: Half duplex switched Ethernet media access latency 1768 ------------------------------------------ 1769 Type Speed Max Pkt Max Access 1770 Length Latency 1771 ------------------------------------------ 1772 Ethernet 10 Mbps 1.2 ms unbounded 1773 100 Mbps 120 us unbounded 1774 1 Gbps 12 us unbounded 1775 ------------------------------------------ 1777 9.4. Half Duplex Switched and Shared Token Ring Networks 1779 In a shared Token Ring network, the network access time for 1780 high priority traffic at any station is bounded and is given 1781 by (N+1)*THTmax, where N is the number of stations sending high 1782 priority traffic and THTmax is the maximum token holding time 1783 [14]. This assumes that network adapters have priority queues 1784 so that reservation of the token is done for traffic with the 1785 highest priority currently queued in the adapter. It is easy to 1786 see that access times can be improved by reducing N or THTmax. The 1787 recommended default for THTmax is 10 ms [6]. N is an integer from 2 1788 to 256 for a shared ring and 2 for a switched half duplex topology. 1789 A similar analysis applies for FDDI. Using the default values gives 1791 Given that access time is bounded, it is possible to provide an 1792 upper bound for end-to-end delays as required by Guaranteed Service 1793 assuming that traffic of this class uses the highest priority 1794 allowable for user traffic. The actual number of stations that send 1795 traffic mapped into the same traffic class as Guaranteed Service may 1796 vary over time but, from an admission control standpoint, this value 1797 Table 7: Half duplex switched and shared Token 1798 Ring media access latency 1799 ---------------------------------------------------- 1800 Type Speed Max Pkt Max Access 1801 Length Latency 1802 ---------------------------------------------------- 1803 Token Ring 4/16 Mbps shared 9 ms 2570 ms 1804 4/16 Mbps switched 9 ms 30 ms 1805 FDDI 100 Mbps 360 us 8 ms 1806 ---------------------------------------------------- 1808 is needed a priori. The admission control entity must therefore use 1809 a fixed value for N, which may be the total number of stations on the 1810 ring or some lower value if it is desired to keep the offered delay 1811 guarantees smaller. If the value of N used is lower than the total 1812 number of stations on the ring, admission control must ensure that 1813 the number of stations sending high priority traffic never exceeds 1814 this number. This approach allows admission control to estimate 1815 worst case access delays assuming that all of the N stations are 1816 sending high priority data even though, in most cases, this will mean 1817 that delays are significantly overestimated. 1819 Assuming that Controlled Load flows use a traffic class lower than 1820 that used by Guaranteed Service, no upper bound on access latency 1821 can be provided for Controlled Load flows. However, Controlled Load 1822 flows will receive better service than best effort flows. 1824 Note that on many existing shared Token Rings, bridges transmit 1825 frames using an Access Priority (see Section 4.3) value of 4 1826 irrespective of the user_priority carried in the frame control 1827 field of the frame. Therefore, existing bridges would need to be 1828 reconfigured or modified before the above access time bounds can 1829 actually be used. 1831 9.5. Half Duplex and Shared Demand Priority Networks 1833 In IEEE 802.12 networks, communication between end nodes and hubs and 1834 between the hubs themselves is based on the exchange of link control 1835 signals. These signals are used to control access to the shared 1836 medium. If a hub, for example, receives a high priority request 1837 while another hub is in the process of serving normal priority 1838 requests, then the service of the latter hub can effectively be 1839 preempted in order to serve the high priority request first. After 1840 the network has processed all high priority requests, it resumes the 1841 normal priority service at the point in the network at which it was 1842 interrupted. 1844 The network access time for high priority packets is basically the 1845 time needed to preempt normal priority network service. This access 1846 time is bounded and it depends on the physical layer and on the 1847 topology of the shared network. The physical layer has a significant 1848 impact when operating in half duplex mode as, e.g. when used across 1849 unshielded twisted pair cabling (UTP) links, because link control 1850 signals cannot be exchanged while a packet is transmitted over the 1851 link. Therefore the network topology has to be considered since, in 1852 larger shared networks, the link control signals must potentially 1853 traverse several links and hubs before they can reach the hub which 1854 has the network control function. This may delay the preemption of 1855 the normal priority service and hence increase the upper bound that 1856 may be guaranteed. 1858 Upper bounds on the high priority access time are given below for a 1859 UTP physical layer and a cable length of 100 m between all end nodes 1860 and hubs using a maximum propagation delay of 570 ns as defined in 1861 [19]. These values consider the worst case signaling overhead and 1862 assume the transmission of maximum sized normal priority data packets 1863 while the normal priority service is being preempted. 1865 Table 8: Half duplex switched Demand Priority UTP access latency 1867 ------------------------------------------------------------ 1868 Type Speed Max Pkt Max Access 1869 Length Latency 1870 ------------------------------------------------------------ 1871 Demand Priority 100 Mbps, 802.3 pkt, UTP 120 us 254 us 1872 802.5 pkt, UTP 360 us 733 us 1873 ------------------------------------------------------------ 1875 Shared IEEE 802.12 topologies can be classified using the hub 1876 cascading level "N". The simplest topology is the single hub network 1877 (N = 1). For a UTP physical layer, a maximum cascading level of 1878 N = 5 is supported by the standard. Large shared networks with 1879 many hundreds of nodes may be built with a level 2 topology. The 1880 bandwidth manager could be informed about the actual cascading level 1881 by network management mechanisms and can use this information in its 1882 admission control algorithms. 1884 In contrast to UTP, the fiber optic physical layer operates in dual 1885 simplex mode. Upper bounds for the high priority access time are 1886 Table 9: Shared Demand Priority UTP access latency 1888 ---------------------------------------------------------------- 1889 Type Speed Max Pkt Max Access Topology 1890 Length Latency 1891 ---------------------------------------------------------------- 1892 Demand Priority 100 Mbps, 802.3 pkt 120 us 262 us N = 1 1893 120 us 554 us N = 2 1894 120 us 878 us N = 3 1895 120 us 1.24 ms N = 4 1896 120 us 1.63 ms N = 5 1898 Demand Priority 100 Mbps, 802.5 pkt 360 us 722 us N = 1 1899 360 us 1.41 ms N = 2 1900 360 us 2.32 ms N = 3 1901 360 us 3.16 ms N = 4 1902 360 us 4.03 ms N = 5 1903 ----------------------------------------------------------------- 1905 given below for 2 km multimode fiber links with a propagation delay 1906 of 10 us. 1908 Table 10: Half duplex switched Demand Priority 1909 fiber access latency 1910 ------------------------------------------------------------ 1911 Type Speed Max Pkt Max Access 1912 Length Latency 1913 ------------------------------------------------------------ 1914 Demand Priority 100 Mbps,802.3 pkt, fiber 120 us 139 us 1915 802.5 pkt, fiber 360 us 379 us 1916 ------------------------------------------------------------ 1918 For shared media with distances of up to 2 km between all end nodes 1919 and hubs, the IEEE 802.12 standard allows a maximum cascading level 1920 of 2. Higher levels of cascaded topologies are supported but require 1921 a reduction of the distances [15]. 1923 The bounded access delay and deterministic network access allow the 1924 support of service commitments required for Guaranteed Service and 1925 Controlled Load, even on shared media topologies. The support of 1926 just two priority levels in 802.12, however, limits the number of 1927 services that can simultaneously be implemented across the network. 1929 Table 11: Shared Demand Priority fiber access latency 1931 --------------------------------------------------------------- 1932 Type Speed Max Pkt Max Access Topology 1933 Length Latency 1934 --------------------------------------------------------------- 1935 Demand Priority 100 Mbps, 802.3 pkt 120 us 160 us N = 1 1936 120 us 202 us N = 2 1938 Demand Priority 100 Mbps, 802.5 pkt 360 us 400 us N = 1 1939 360 us 682 us N = 2 1940 --------------------------------------------------------------- 1942 10. Justification 1944 An obvious concern is the complexity of this model. It essentially 1945 does what RSVP already does at Layer 3, so why do we think we can do 1946 better by reinventing the solution to this problem at Layer 2? 1948 The key is that there are a number of simple Layer 2 scenarios 1949 that cover a considerable portion of the real QoS problems that 1950 will occur. A solution that covers the majority of problems at 1951 significantly lower cost is beneficial. Full RSVP/IntServ with per 1952 flow queueing in strategically positioned high function switches or 1953 routers may be needed to completely resolve all issues, but devices 1954 implementing the architecture described in herein will allow for a 1955 significantly simpler network. 1957 11. Summary 1959 This document has specified a framework for providing Integrated 1960 Services over shared and switched LAN technologies. The ability to 1961 provide QoS guarantees necessitates some form of admission control 1962 and resource management. The requirements and goals of a resource 1963 management scheme for subnets have been identified and discussed. 1964 We refer to the entire resource management scheme as a Bandwidth 1965 Manager. Architectural considerations were discussed and examples 1966 were provided to illustrate possible implementations of a Bandwidth 1967 Manager. Some of the issues involved in mapping the services 1968 from higher layers to the link layer have also been discussed. 1969 Accompanying memos from the ISSLL working group address service 1970 mapping issues [13] and provide a protocol specification for the 1971 Bandwidth Manager protocol [14] based on the requirements and goals 1972 discussed in this document. 1974 References 1976 [1] IEEE Standards for Local and Metropolitan Area Networks: Overview 1977 and Architecture, ANSI/IEEE Std 802, 1990. 1979 [2] ISO/IEC 10038 Information technology - Telecommunications and 1980 information exchange between systems - Local area networks - Media 1981 Access Control (MAC) Bridges, (also ANSI/IEEE Std 802.1D-1993), 1982 1993. 1984 [3] ISO/IEC Final CD 15802-3 Information technology - Tele- 1985 communications and information exchange between systems - 1986 Local and metropolitan area networks - Common specifications - 1987 Part 3: Media Access Control (MAC) bridges, (current draft 1988 available as IEEE P802.1D/D15). 1990 [4] IEEE Standards for Local and Metropolitan Area Networks: Draft 1991 Standard for Virtual Bridged Local Area Networks, P802.1Q/D8, 1992 January 1998. 1994 [5] B. Braden, L. Zhang, S. Berson, S. Herzog and S. Jamin, Resource 1995 Reservation Protocol (RSVP) - Version 1 Functional Specification, 1996 RFC 2205, September 1997. 1998 [6] J. Wroclawski, Specification of the Controlled Load Network Element 1999 Service, RFC 2211, September 1997. 2001 [7] S. Shenker, C. Partridge and R. Guerin, Specification of Guaranteed 2002 Quality of Service, RFC 2212, September 1997. 2004 [8] R. Braden, D. Clark and S. Shenker, Integrated Services in the 2005 Internet Architecture: An Overview, RFC 1633, June 1994. 2007 [9] J. Wroclawski, The Use of RSVP with IETF Integrated Services, 2008 RFC 2210, September 1997. 2010 [10] S. Shenker and J. Wroclawski, Network Element Service Specification 2011 Template, RFC 2216, September 1997. 2013 [11] S. Shenker and J. Wroclawski, General Characterization Parameters 2014 for Integrated Service Network Elements, RFC 2215, September 1997. 2016 [12] L. Delgrossi and L. Berger (Editors), Internet Stream Protocol 2017 Version 2 (ST2) Protocol Specification - Version ST2+, 2018 RFC 1819, August 1995. 2020 [13] M. Seaman, A. Smith, and E. Crawley, Integrated Service Mappings on 2021 IEEE 802 Networks, work in progress, Internet Draft, 2022 , November 1997. 2024 [14] R. Yavatkar, D. Hoffman, Y. Bernet, and F. Baker, SBM 2025 (Subnet Bandwidth Manager): Protocol for RSVP-based Admission 2026 Control over IEEE 802-style networks, work in progress, 2027 Internet Draft, , 2028 November 1997. 2030 [15] ISO/IEC 8802-3 Information technology - Telecommunications and 2031 information exchange between systems - Local and metropolitan 2032 area networks - Common specifications - Part 3: Carrier Sense 2033 Multiple Access with Collision Detection (CSMA/CD) Access Method 2034 and Physical Layer Specifications, (also ANSI/IEEE Std 802.3-1996), 2035 1996. 2037 [15] ISO/IEC 8802-5 Information technology - Telecommunications and 2038 information exchange between systems - Local and metropolitan 2039 area networks - Common specifications - Part 5: Token Ring Access 2040 Method and Physical Layer Specifications, (also 2041 ANSI/IEEE Std 802.5-1995), 1995. 2043 [17] J. Postel and J. Reynolds, A Standard for the Transmission of 2044 IP Datagrams over IEEE 802 Networks, RFC 1042, February 1988. 2046 [18] C. Bisdikian, B. V. Patel, F. Schaffa, and M Willebeek-LeMair, 2047 The Use of Priorities on Token Ring Networks for Multimedia 2048 Traffic, IEEE Network, Nov/Dec 1995. 2050 [19] IEEE Standards for Local and Metropolitan Area Networks: 2051 Demand Priority Access Method, Physical Layer and Repeater 2052 Specification for 100 Mb/s Operation, IEEE Std 802.12-1995. 2054 [20] Fiber Distributed Data Interface MAC, ANSI Std. X3.139-1987. 2056 Security Considerations 2058 Implementation of the model described in this memo creates no known 2059 new avenues for malicious attack on the network infrastructure 2060 although readers are referred to Section 2.8 of the RSVP 2061 specification [5] for a discussion of the impact of the use of 2062 admission control signaling protocols on network security. 2064 Acknowledgements 2066 Much of the work presented in this document has benefited greatly 2067 from discussion held at the meetings of the Integrated Services 2068 over Specific Link Layers (ISSLL) working group. We would like to 2069 acknowledge contributions from the many participants via discussion 2070 at these meetings and on the mailing list. We would especially like 2071 to thank Eric Crawley, Don Hoffman and Raj Yavatkar for contributions 2072 via previous Internet drafts, and Peter Kim for contributing the text 2073 about Demand Priority networks. 2075 Authors' Addresses 2077 Anoop Ghanwani 2078 Nortel Networks 2079 3 Federal St, BL3-03 2080 Billerica, MA 01821, USA 2081 +1-978-916-4514 2082 aghanwan@nortelnetworks.com 2084 J. Wayne Pace 2085 IBM Corporation 2086 P. O. Box 12195 2087 Research Triangle Park, NC 27709, USA 2088 +1-919-254-4930 2089 pacew@raleigh.ibm.com 2091 Vijay Srinivasan 2092 Torrent Networking Technologies 2093 3000 Aerial Center Parkway, Ste 140 2094 Morrisville, NC 27560, USA 2095 +1-919-468-8466 2096 vijay@torrentnet.com 2098 Andrew Smith 2099 Extreme Networks 2100 3585 Monroe St 2101 Santa Clara, CA 95051, USA 2102 +1-408-579-2821 2103 andrew@extremenetworks.com 2105 Mick Seaman 2106 3Com Corporation 2107 5400 Bayfront Plaza 2108 Santa Clara CA 95052, USA 2109 +1-408-764-5000 2110 mick_seaman@3com.com