|
|
This website is designed for a Packet Classification Cache Simulator (PCCS).
Packet classification is that process that determines which flow a packet belongs to based on one or more fields in the packet. Typical classifications being done today involve fields of a packet's header including the source IP address, the source port, destination IP address, destination port, and protocol. With the increasing speeds of modern networks, the decreasing packet sizes of emerging network applications and the economic realities facing equipment vendors, it is of paramount importance to deliver fast packet classification functionality at a very low cost. It has been well-established that memory access delays limit the classification speeds. While the lookup algorithm itself can be implemented in hardware, the dynamic nature of the classifying rules requires that the classification table be stored in memory. As memory speeds have not kept pace with the rest of the hardware advances, classification speeds are limited by memory access latency. Unfortunately, the best solutions to this problem still require a significantly higher number of memory accesses. The best way to speed this classification lookup is to avoid doing it by caching previous classification decisions and using them directly. Caching improves lookup speeds by taking advantage of the locality in the traffic. While full classification algorithms require multiple memory accesses, cache lookups can be performed using a single memory access. As the PC CPU-memory cache, many aspects of cache, such as associativities, hash functions, and replacement policies, require to be tuned to achieve high performance. We build PCCS for the purpose of tuning these aspects. The goal is to design a packet classification cache with very high hit rate for various realistic traffic, while use as less memory as possible. To learn more about the project, developers please go to the project page. To download the software, please go to the project CVS page |
|
|
|
|
|
Traffic Traces:
|
|
|||
Kang Li | Francis Chang | Damien Burger | Wu-chang Feng |
Last update: 12-Feb-2003 |