Efficient prefix cache for network processors

Abstract

Abstract-Conventional routing cache systems store destination IP addresses in their cache directory. We present a routing cache technique that stores the most recently used route prefixes, instead of IP addresses, to achieve significantly smaller cache size. A nesting prefix is partially represented in this cache by its minimal expansions. Such expanded prefixes are obtained using an incremental technique without any modifications to the routing table. Consequently, our cache works with most of the common route lookup algorithms and efficiently maintains coherency with the routing table. Experiments show that for a hit ratio over 0.96 our design can achieve more than 33 times reduction in cache size, compared to a conventional routing cache.

PC-DUOS: Fast TCAM lookup and update for packet classifiers

“…Since the LTCAM stores independent rules, at most one rule may match during lookup in the LTCAM and a priority encoder is not needed. If a match is found in the LTCAM during lookup, it is guaranteed to be the highest priority match and the corresponding action can be returned immediately yielding a 50% [1] improvement in TCAM search time. The average improvement in lookup time is estimated to be between 20% to 48% for the tests in our data set.…”

“…For a single lookup finding a match in the LTCAM, the improvement is 50% [1]. The best case in our dataset is test1, which has 98% of the rules in the LTCAM.…”

“…At the same time the ongoing lookup process on the ITCAM (which takes longer due to the priority resolution step) is aborted. Thus, if a match is found on the LTCAM, the overall lookup time is shortened by about 50% [1]. The final stage logic in Figure 1 that chooses between the two next hops could be moved ahead and placed between the TCAM and SRAM stages.…”

“…For example, the packet could be forwarded to an appropriate output link, or it may be dropped. A d-dimensional filter F is a dtuple (F [1], F [2], · · · , F [d]), where F [i] is a range specified for an attribute in the packet header, such as destination address, source address, port number, protocol type, TCP flag, etc. A packet matches filter F , if its attribute values fall in the ranges of F [1], · · · , F [d].…”

“…A d-dimensional filter F is a dtuple (F [1], F [2], · · · , F [d]), where F [i] is a range specified for an attribute in the packet header, such as destination address, source address, port number, protocol type, TCP flag, etc. A packet matches filter F , if its attribute values fall in the ranges of F [1], · · · , F [d]. Since it is possible for a packet to match more than one of the filters in a classifier thereby resulting in a tie, each rule has an associated cost or priority.…”