Digest caches have been proposed as an effective method to speed up packet classification in network processors. In this paper, we show that the presence of a large number of small flows and a few large flows in the Internet has an adverse impact on the performance of these digest caches. In the Internet, a few large flows transfer a majority of the packets whereas the contribution of several small flows to the total number of packets transferred is small. In such a scenario, the LRU cache replacement policy, which gives maximum priority to the most recently accessed digest, tends to evict digests belonging to the few large flows. We propose a new cache management algorithm called Saturating Priority (SP) which aims at improving the performance of digest caches in network processors by exploiting the disparity between the number of flows and the number of packets transferred. Our experimental results demonstrate that SP performs better than the widely used LRU cache replacement policy in size constrained caches. Further, we characterize the misses experienced by flow identifiers in digest caches. © 2008 Springer Berlin Heidelberg.
CITATION STYLE
Chandramohan, G., & Ramaswamy, G. (2008). Improving performance of digest caches in network processors. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 5374 LNCS, pp. 6–17). Springer Verlag. https://doi.org/10.1007/978-3-540-89894-8_5
Mendeley helps you to discover research relevant for your work.