Scalable Many-Field Packet Classification on Multi-core Processors

Abstract

Packet classification matches a packet header against the predefined rules in a rule set; it is a kernel function that has been studied for decades. A recent trend in packet classification is to match a large number of packet header fields. For example, the flow table lookup in Software Defined Networking (SDN) requires 15 fields of the packet header to be examined. Another trend in packet classification is to use software-based solutions employing multi-core general purpose processors and virtual machines. Although packet classification has been widely studied, most existing solutions on multi-core systems target the classic 5-field packet classification; their performance cannot be easily scaled up for a larger number of packet header fields. In this paper, we propose a decompositionbased packet classification approach; it supports large rule sets consisting of a large number of packet header fields. We first use range-tree and hashing to search each field of the input packet header individually in parallel. The partial results from all the fields are represented by bit vectors; they are merged in parallel to produce the final packet header match. We also balance the search and merge latencies, and employ software pipelining to further enhance the overall performance. We implement our approach on state-of-the-art multi-core processors; we evaluate its performance with respect to throughput and latency for rule set size ranging from 1K to 32K. Experimental results show that, for a 32K rule set, our algorithms can achieve an average processing latency of 2000 ns per packet and an overall throughput of 30 million packets per second on a state-of-the-art 16-core platform.

Performance modeling and optimizations for decomposition-based large-scale packet classification on multi-core processors

“…Bit Vector (BV) [4] is a data structure used for merging the partial matching results. For a rule set consisting of N rules, a partial matching result from a specific field is a BV of N bits, each bit corresponding to a particular rule in the III.…”

“…, M − 1; (2) use a hash value i as the index to access the hash table T and choose another set of M hash functions. We do not introduce the range-tree and the hashing mechanism in detail, since they have been extensively studied [4], [20].…”

“…We use the same hash-based search techniques as in [4]. Note only one BV is extracted for each field in the search phase.…”

“…To implement our packet classification engine, we can assign each core a single packet header field; however, the communication overhead between different cores during the merge phase limits the performance of this implementation [4]. Therefore, in our implementation, each packet header of W -fields stays in a single core.…”

“…As shown in Table III, Vprty (3-bit), MPLS tfc (3-bit) and ToS fields (6-bit) are short fields, the maximum number of unique rules in these fields is very small; therefore the numbers of unique rules in these fields are saturated (ρ (k) = 1). Increasing the number of unique rules is similar to enlarging the original rule set [4], which has a negative effect on the performance (Section V-C). To reduce the number of unique rules after grouping, we only group those (short fields) containing a small number of bits: Eth type, MPLS lbl, and ToS are grouped as one superfield, while Ingr, VID, Vprty, Prtl, and MPLS tfc are grouped as another superfield.…”