idnits 2.17.1 draft-mcbride-armd-mcast-overview-00.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- == It seems as if not all pages are separated by form feeds - found 0 form feeds but 10 pages Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- == There are 2 instances of lines with multicast IPv4 addresses in the document. If these are generic example addresses, they should be changed to use the 233.252.0.x range defined in RFC 5771 Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (March 3, 2012) is 4437 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- -- Obsolete informational reference (is this intentional?): RFC 4601 (Obsoleted by RFC 7761) Summary: 0 errors (**), 0 flaws (~~), 3 warnings (==), 2 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 1 Internet Engineering Task Force M. McBride 2 Internet-Draft H. Lui 3 Intended status: Informational Huawei Technologies 4 Expires: September 4, 2012 March 3, 2012 6 Multicast in the Data Center Overview 7 draft-mcbride-armd-mcast-overview-00 9 Abstract 11 There has been much interest in issues surrounding massive amounts of 12 hosts in the data center. There was a discussion, in ARMD, involving 13 the issues with address resolution for non ARP/ND multicast traffic 14 in data centers with massive number of hosts. This document provides 15 a quick survey of multicast in the data center and should serve as an 16 aid to further discussion of issues related to large amounts of 17 multicast in the data center. 19 Status of this Memo 21 This Internet-Draft is submitted in full conformance with the 22 provisions of BCP 78 and BCP 79. 24 Internet-Drafts are working documents of the Internet Engineering 25 Task Force (IETF). Note that other groups may also distribute 26 working documents as Internet-Drafts. The list of current Internet- 27 Drafts is at http://datatracker.ietf.org/drafts/current/. 29 Internet-Drafts are draft documents valid for a maximum of six months 30 and may be updated, replaced, or obsoleted by other documents at any 31 time. It is inappropriate to use Internet-Drafts as reference 32 material or to cite them other than as "work in progress." 34 This Internet-Draft will expire on September 4, 2012. 36 Copyright Notice 38 Copyright (c) 2012 IETF Trust and the persons identified as the 39 document authors. All rights reserved. 41 This document is subject to BCP 78 and the IETF Trust's Legal 42 Provisions Relating to IETF Documents 43 (http://trustee.ietf.org/license-info) in effect on the date of 44 publication of this document. Please review these documents 45 carefully, as they describe your rights and restrictions with respect 46 to this document. Code Components extracted from this document must 47 include Simplified BSD License text as described in Section 4.e of 48 the Trust Legal Provisions and are provided without warranty as 49 described in the Simplified BSD License. 51 Table of Contents 53 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3 54 2. Multicast Applications in the Data Center . . . . . . . . . . . 3 55 2.1. L3 Multicast Applications . . . . . . . . . . . . . . . . . 3 56 2.2. L2 Multicast Applications . . . . . . . . . . . . . . . . . 4 57 3. L2 Multicast Protocols in the Data Center . . . . . . . . . . . 5 58 4. L3 Multicast solutions in the Data Center . . . . . . . . . . . 6 59 5. Challenges of using multicast in the Data Center . . . . . . . 7 60 6. Layer 3 / Layer 2 Topological Variations . . . . . . . . . . . 8 61 7. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 9 62 8. IANA Considerations . . . . . . . . . . . . . . . . . . . . . . 9 63 9. Security Considerations . . . . . . . . . . . . . . . . . . . . 9 64 10. Informative References . . . . . . . . . . . . . . . . . . . . 9 65 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 9 67 1. Introduction 69 Data center servers often use IP Multicast to send data to clients or 70 other application servers. IP Multicast is expected to help conserve 71 bandwidth in the data center and reduce the load on servers. 72 Increased reliance on multicast, in next generation data centers, 73 requires higher performance and capacity especially from the 74 switches. If multicast is to continue to be used in the data center, 75 it must scale well within and between datacenters. There has been 76 much interest in issues surrounding massive amounts of hosts in the 77 data center. There was a discussion, in ARMD, involving the issues 78 with address resolution for non ARP/ND multicast traffic in data 79 centers. This document provides a quick survey of multicast in the 80 data center and should serve as an aid to further discussion of 81 issues related to multicast in the data center. 83 ARP/ND issues are not addressed in this document. ARP/ND issues are 84 addressed in [I-D.armd-problem-statement] 86 2. Multicast Applications in the Data Center 88 There are many data center operators who do not deploy Multicast in 89 their networks for scalability and stability reasons. There are also 90 many operators for whom multicast is critical and is enabled on their 91 data center switches and routers. For this latter group, there are 92 several uses of multicast in their data centers. An understanding of 93 the uses of that multicast is important in order to properly support 94 these applications in the ever evolving data centers. If, for 95 instance, the majority of the applications are discovering/signaling 96 each other using multicast there may be better ways to support them 97 then using multicast. If, however, the multicasting of data is 98 occurring in large volumes, there is a need for very good data center 99 under/overlay multicast support. The applications either fall into 100 the category of those that leverage L2 multicast for discovery or of 101 those that require L3 support and likely span multiple subnets. 103 2.1. L3 Multicast Applications 105 IPTV servers use multicast to deliver content from the data center to 106 end users. IPTV is typically a one to many application where the 107 hosts are configured for IGMPv3, the switches are configured with 108 IGMP snooping, and the routers are running PIM-SSM mode. Often 109 redundant servers are sending multicast streams into the network and 110 the network is forwarding the data across diverse paths. 112 Windows Media servers send multicast streaming to clients. Windows 113 Media Services streams to an IP multicast address and all clients 114 subscribe to the IP address to receive the same stream. This allows 115 a single stream to be played simultaneously by multiple clients and 116 thus reducing bandwidth utilization. 118 Market data relies extensively on IP multicast to deliver stock 119 quotes from the data center to a financial services provider and then 120 to the stock analysts. The most critical requirement of a multicast 121 trading floor is that it be highly available. The network must be 122 designed with no single point of failure and in a way the network can 123 respond in a deterministic manner to any failure. Typically 124 redundant servers (in a primary/backup or live live mode) are sending 125 multicast streams into the network and the network is forwarding the 126 data across diverse paths (when duplicate data is sent by multiple 127 servers). 129 With publish and subscribe servers a separate message is sent to each 130 subscriber of a publication. With multicast publish/subscribe, only 131 one message is sent, regardless of the number of subscribers. In a 132 publish/subscribe system, client applications, some of which are 133 publishers and some of which are subscribers, are connected to a 134 network of message brokers that receive publications on a number of 135 topics, and send the publications on to the subscribers for those 136 topics. The more subscribers there are in the publish/subscribe 137 system, the greater the improvement to network utilization there 138 might be with multicast. 140 With load balancing protocols, such as VRRP, routers communicate 141 within themselves using a multicast address. 143 Overlays may use IP multicast to virtualize L2 multicasts. VXLAN, 144 for instance, is an encapsulation scheme to carry L2 frames over L3 145 networks. The VXLAN Tunnel End Point (VTEP) encapsulates frames 146 inside an L3 tunnel. VXLANs are identified by a 24 bit VXLAN Network 147 Identifier (VNI). The VTEP maintains a table of known destination 148 MAC addresses, and stores the IP address of the tunnel to the remote 149 VTEP to use for each. Unicast frames, between VMs, are sent directly 150 to the unicast L3 address of the remote VTEP. Multicast frames are 151 sent to a multicast IP group associated with the VNI. Underlying IP 152 Multicast protocols (PIM-SM/SSM/BIDIR) are used to forward multicast 153 data across the overlay. 155 2.2. L2 Multicast Applications 157 Applications, such as Ganglia, uses multicast for distributed 158 monitoring of computing systems such as clusters and grids. 160 Windows Server, cluster node exchange, relies upon the use of 161 multicast heartbeats between servers. Only the other interfaces in 162 the same multicast group use the data. Unlike broadcast, multicast 163 traffic does not need to be flooded throughout the network, reducing 164 the chance that unnecessary CPU cycles are expended filtering traffic 165 on nodes outside the cluster. As the number of nodes increases, the 166 ability to replace several unicast messages with a single multicast 167 message improves node performance and decreases network bandwidth 168 consumption. Multicast messages replace unicast messages in two 169 components of clustering: 171 o Heartbeats: The clustering failure detection engine is based on a 172 scheme whereby nodes send heartbeat messages to other nodes. 173 Specifically, for each network interface, a node sends a heartbeat 174 message to all other nodes with interfaces on that network. 175 Heartbeat messages are sent every 1.2 seconds. In the common case 176 where each node has an interface on each cluster network, there 177 are N * (N - 1) unicast heartbeats sent per network every 1.2 178 seconds in an N-node cluster. With multicast heartbeats, the 179 message count drops to N multicast heartbeats per network every 180 1.2 seconds, because each node sends 1 message instead of N - 1. 181 This represents a reduction in processing cycles on the sending 182 node and a reduction in network bandwidth consumed. 184 o Regroup: The clustering membership engine executes a regroup 185 protocol during a membership view change. The regroup protocol 186 algorithm assumes the ability to broadcast messages to all cluster 187 nodes. To avoid unnecessary network flooding and to properly 188 authenticate messages, the broadcast primitive is implemented by a 189 sequence of unicast messages. Converting the unicast messages to 190 a single multicast message conserves processing power on the 191 sending node and reduces network bandwidth consumption. 193 Multicast addresses in the 224.0.0.x range are considered link local 194 multicast addresses. They are used for protocol discovery and are 195 flooded to every port. For example, OSPF uses 224.0.0.5 and 196 224.0.0.6 for neighbor and DR discovery. These addresses are 197 reserved and will not be constrained by IGMP snooping. These 198 addresses are not to be used by any application. 200 These types of multicast applications should be able to be supported 201 in data centers which support multicast. 203 3. L2 Multicast Protocols in the Data Center 205 The switches, in between the servers and the routers, rely upon igmp 206 snooping to bound the multicast to the ports leading to interested 207 hosts and to L3 routers. A switch will, by default, flood multicast 208 traffic to all the ports in a broadcast domain (VLAN). IGMP snooping 209 is designed to prevent hosts on a local network from receiving 210 traffic for a multicast group they have not explicitly joined. It 211 provides switches with a mechanism to prune multicast traffic from 212 links that do not contain a multicast listener (an IGMP client). 213 IGMP snooping is a L2 optimization for L3 IGMP. 215 IGMP snooping, with proxy reporting or report suppression, actively 216 filters IGMP packets in order to reduce load on the multicast router. 217 Joins and leaves heading upstream to the router are filtered so that 218 only the minimal quantity of information is sent. The switch is 219 trying to ensure the router only has a single entry for the group, 220 regardless of how many active listeners there are. If there are two 221 active listeners in a group and the first one leaves, then the switch 222 determines that the router does not need this information since it 223 does not affect the status of the group from the router's point of 224 view. However the next time there is a routine query from the router 225 the switch will forward the reply from the remaining host, to prevent 226 the router from believing there are no active listeners. It follows 227 that in active IGMP snooping, the router will generally only know 228 about the most recently joined member of the group. 230 In order for IGMP, and thus IGMP snooping, to function, a multicast 231 router must exist on the network and generate IGMP queries. The 232 tables (holding the member ports for each multicast group) created 233 for snooping are associated with the querier. Without a querier the 234 tables are not created and snooping will not work. Furthermore IGMP 235 general queries must be unconditionally forwarded by all switches 236 involved in IGMP snooping. Some IGMP snooping implementations 237 include full querier capability. Others are able to proxy and 238 retransmit queries from the multicast router. 240 In source-only networks, however, which presumably describes most 241 data center networks, there are no IGMP hosts on switch ports to 242 generate IGMP packets. Switch ports are connected to multicast 243 source ports and multicast router ports. The switch typically learns 244 about multicast groups from the multicast data stream by using a type 245 of source only learning (when only receiving multicast data on the 246 port, no IGMP packets). The switch forwards traffic only to the 247 multicast router ports. When the switch receives traffic for new IP 248 multicast groups, it will typically flood the packets to all ports in 249 the same VLAN. This unnecessary flooding can impact switch 250 performance. 252 4. L3 Multicast solutions in the Data Center 254 There are three flavors of PIM used for Multicast Routing in the Data 255 Center: PIM-SM [RFC4601], PIM-SSM [RFC4607], and PIM-BIDIR [RFC5015]. 257 SSM provides the most efficient forwarding between sources and 258 receivers and is most suitable for one to many types of multicast 259 applications. State is built for each S,G channel therefore the more 260 sources and groups there are, the more state there is in the network. 261 BIDIR is the most efficient shared tree solution as one tree is built 262 for all S,G's, therefore saving state. But it is not the most 263 efficient in forwarding path between sources and receivers. SSM and 264 BIDIR are optimizations of PIM-SM. PIM-SM is still the most widely 265 deployed multicast routing protocol. PIM-SM can also be the most 266 complex. PIM-SM relies upon a RP (Rendezvous Point) to set up the 267 multicast tree and then will either switch to the SPT (shortest path 268 tree), similar to SSM, or stay on the shared tree (similar to BIDIR). 269 For massive amounts of hosts sending (and receiving) multicast, the 270 shared tree (particularly with PIM-BIDIR) provides the best potential 271 scaling since no matter how many multicast sources exist within a 272 VLAN, the tree number stays the same. IGMP snooping, IGMP proxy, and 273 PIM-BIDIR have the potential to scale to the huge scaling numbers 274 required in a data center. 276 5. Challenges of using multicast in the Data Center 278 When IGMP/MLD Snooping is not implemented, ethernet switches will 279 flood multicast frames out of all switch-ports, which turns the 280 traffic into something more like broadcast. 282 VRRP uses multicast heartbeat to communicate between routers. The 283 communication between the host and the default gateway is unicast. 284 The multicast heartbeat can be very chatty when there are thousands 285 of VRRP pairs with sub-second heartbeat calls back and forth. 287 Link-local multicast should scale well within one IP subnet 288 particularly with a large layer3 domain extending down to the access 289 or aggregation switches. But if multicast traverses beyond one IP 290 subnet, which is necessary for an overlay like VXLAN, you could 291 potentially have scaling concerns. If using a VXLAN overlay, it is 292 necessary to map the L2 multicast in the overlay to L3 multicast in 293 the underlay or do head end replication in the overlay and receive 294 duplicate frames on the first link from the router to the core 295 switch. The solution could be to run potentially thousands of PIM 296 messages to generate/maintain the required multicast state in the IP 297 underlay. The behavior of the upper layer, with respect to 298 broadcast/multicast, affects the choice of head end (*,G) or (S,G) 299 replication in the underlay, which affects the opex and capex of the 300 entire solution. A VXLAN, with thousands of logical groups, maps to 301 head end replication in the hypervisor or to IGMP from the hypervisor 302 and then PIM between the TOR and CORE 'switches' and the gateway 303 router. 305 Requiring IP multicast (especially PIM BIDIR) from the network can 306 prove challenging for data center operators especially at the kind of 307 scale that the VXLAN/NVGRE proposals require. This is also true when 308 the L2 topological domain is large and extended all the way to the L3 309 core. In data centers with highly virtualized servers, even small L2 310 domains may spread across many server racks (i.e. multiple switches 311 and router ports). 313 It's not uncommon for there to be 10-20 VMs per server in a 314 virtualized environment. One vendor reported a customer requesting a 315 scale to 400VM's per server. For multicast to be a viable solution 316 in this environment, the network needs to be able to scale to these 317 numbers when these VMs are sending/receiving multicast. 319 A lot of switching/routing hardware has problems with IP Multicast, 320 particularly with regards to hardware support of PIM-BIDIR. 322 Sending L2 multicast over a campus or data center backbone, in any 323 sort of significant way, is a new challenge enabled for the first 324 time by overlays. There are interesting challenges when pushing 325 large amounts of multicast traffic through a network, and have thus 326 far been dealt with using purpose-built networks. While the overlay 327 proposals have been careful not to impose new protocol requirements, 328 they have not addressed the issues of performance and scalability, 329 nor the large-scale availability of these protocols. 331 There is an unnecessary multicast stream flooding problem in the link 332 layer switches between the multicast source and the PIM First Hop 333 Router (FHR). The IGMP-Snooping Switch will forward multicast 334 streams to router ports, and the PIM FHR must receive all multicast 335 streams even if there is no request from receiver. This often leads 336 to waste of switch cache and link bandwidth when the multicast 337 streams are not actually required. [I-D.pim-umf-problem-statement] 338 details the problem and defines design goals for a generic mechanism 339 to restrain the unnecessary multicast stream flooding. 341 6. Layer 3 / Layer 2 Topological Variations 343 As discussed in [I-D.armd-problem-statement], there are a variety of 344 topological data center variations including L3 to Access Switches, 345 L3 to Aggregation Switches, and L3 in the Core only. Further 346 analysis is needed in order to understand how these variations affect 347 IP Multicast scalability 349 7. Acknowledgements 351 The authors would like to thank the many individuals who contributed 352 opinions on the ARMD wg mailing list about this topic: Linda Dunbar, 353 Anoop Ghanwani, Peter Ashwoodsmith, David Allan, Aldrin Isaac, Igor 354 Gashinsky, Michael Smith, Patrick Frejborg, Joel Jaeggli and Thomas 355 Narten. 357 8. IANA Considerations 359 This memo includes no request to IANA. 361 9. Security Considerations 363 No security considerations at this time. 365 10. Informative References 367 [I-D.armd-problem-statement] 368 Narten, T., Karir, M., and I. Foo, 369 "draft-ietf-armd-problem-statement", February 2012. 371 [I-D.pim-umf-problem-statement] 372 Zhou, D., Deng, H., Shi, Y., Liu, H., and I. Bhattacharya, 373 "draft-dizhou-pim-umf-problem-statement", October 2010. 375 [RFC4601] Fenner, B., Handley, M., Holbrook, H., and I. Kouvelas, 376 "Protocol Independent Multicast - Sparse Mode (PIM-SM): 377 Protocol Specification (Revised)", RFC 4601, August 2006. 379 [RFC4607] Holbrook, H. and B. Cain, "Source-Specific Multicast for 380 IP", RFC 4607, August 2006. 382 [RFC5015] Handley, M., Kouvelas, I., Speakman, T., and L. Vicisano, 383 "Bidirectional Protocol Independent Multicast (BIDIR- 384 PIM)", RFC 5015, October 2007. 386 Authors' Addresses 388 Mike McBride 389 Huawei Technologies 390 2330 Central Expressway 391 Santa Clara, CA 95050 392 USA 394 Email: michael.mcbride@huawei.com 396 Helen Lui 397 Huawei Technologies 398 Building Q14, No. 156, Beiqing Rd. 399 Beijing, 100095 400 China 402 Email: helen.liu@huawei.com