Top Document: comp.dcom.sys.cisco Frequently Asked Questions (FAQ) Previous Document: How do cisco routers rate performance-wise? Next Document: How does one interpret buffer statistics? See reader questions & answers on this topic! - Help others by sharing your knowledge There are 3 basic types of switching (in order of increasing performance). process switching fast switching autonomous switching Process and fast switching support inbound and outbound, simple and extended, access lists. Of course, for fast switching, such lists only restrict traffic on the particular fast-switched interface. Autonomous switching is done in the switch processor, a microcoded device that is capable of switching IP, IPX, and bridging packets in the 100kpps range. This is known as the "SP" card on the 7000 and the CBUS controller on the AGS+. Encapsulation support is rather limited (Ethernet, HDLC, HSSI...). The cisco 7000 also supports: silicon switching Silicon switching is done in the silicon switching engine (creative, eh? ;-). The silicon switch processor (SSP) is the board which combines both the switch processor and a silicon switching engine. The SSP supports simple and extended outbound access lists in 10.3 and later. The SSP supports simple and extended inbound access lists in 11.1 and later. The cisco 75xx series supports: "optimal" switching (cruddy name, eh?) "flow" switching "distributed" switching * "optimal" switching (cruddy name, eh?) The 7500 platform does not have a separate SP or SSP card, rather the RISC processor on the "integrated route/switch processor card (IRSP)" handles switching directly, similar to the 4000 series routers. There are several hardware and software enhancements made though to increase the throughput to a level that is several times above what you would normally get from "fast" switching. Everything that "fast" switching supports is supported in "optimal" switching. * "flow" switching Basicly the "optimal" switching method, however things have been front-ended with an additional small "flow" cache. This flow cache contains information about source/destination addresses & ports which allow the router to make more informed queueing decisions and process access lists faster. This is a win in routers that would tend to carry a reasonably small number of flows at any one time, such as what you would expect in a corporate network or in a smaller internet service provider network. It's unclear if there are any advantages in a large internet backbone. * "distributed" switching cisco has announced a new type of interface-processor card, called a "VIP" available in the 7500 platform that is intelligent enough to switch packets with no intervention on the part of the IRSP card. This once again separates switching from routing, as in the earlier CBUS/SP/SSP design. The first packet of every session or connection is always Process Switched. The route table is consulted (this resides in DRAM on the CPU) and the "result" is cached in the system memory cache. If the protocol can only be process switched, then it will continue this way and interrupt the CPU for a route table lookup each time. [comment: Process Switching is brutally slow compared to other switching methods. Some features (usually new features do this for the first few software releases) force every packet to be process switched. If you can't avoid process-switching every packet, at least get a router with a fast CPU, such as the 75xx, 4500, and 4700. The 4700 is currently the fastest at process-switching packets, with the 4500 and 75xx tied for second. The 75xx can optimum-switch, however, so it's a lot faster than either of the 4x00s, if you can use it). The second and subsequent packets of each session are capable of being Fast Switched (more session types are becoming fast-switchable), and will consult only the route-cache. This still involves a memory lookup on the board, but the packet can be transferred from the source card directly to the destination card without requiring full storage on the CSC [the CSC refers to the CPU card, basically]. There are some undocumented commands that are useful for obtaining per-interface statistics on what sort of switching was performed. For instance: frobozz-magic-robot>sh int atm4/0 switch ATM4/0 Throttle count: 0 Protocol Path Pkts In Chars In Pkts Out Chars Out IP Process 104851 7669968 116378 11180988 Cache misses 35826 Fast 0 0 0 0 Auton/SSE 0 0 0 0 frobozz-magic-robot>sh int atm4/0 stat ATM4/0 Switching path Pkts In Chars In Pkts Out Chars Out Processor 105024 7679155 116422 11184108 Route cache/FIB 0 0 0 0 Distributed cache 0 0 0 0 Total 105024 7679155 116422 11184108 User Contributions:Top Document: comp.dcom.sys.cisco Frequently Asked Questions (FAQ) Previous Document: How do cisco routers rate performance-wise? Next Document: How does one interpret buffer statistics? Single Page [ Usenet FAQs | Web FAQs | Documents | RFC Index ] Send corrections/additions to the FAQ Maintainer: cisco-faq@panix.com (comp.dcom.sys.cisco FAQ responses)
Last Update March 27 2014 @ 02:11 PM
|
Comment about this article, ask questions, or add new information about this topic: