Scaling multi-core network processors without the reordering bottleneck

Abstract

Today, designers of network processors strive to keep the packet reception and transmission orders identical, and therefore avoid any possible out-of-order transmission. However, the development of new features in advanced network processors has resulted in increasingly parallel architectures and increasingly heterogeneous packet processing times, leading to large reordering delays. In this paper, we introduce novel scalable scheduling algorithms for preserving flow order in parallel multi-core network processors. We show how these algorithms can reduce reordering delay while adapting to any load-balancing algorithm and keeping a low implementation complexity overhead. To do so, we use the observation that all packets in a given flow have similar processing requirements and can be described with a constant number of logical processing phases. We further define three possible knowledge frameworks of the time when a network processor learns about these logical phases, and deduce appropriate algorithms for each of these frameworks.

Scaling Multi-Core Network Processors without the Reordering Bottleneck

“…As mentioned, in our online technical report [13], we further discuss the implementation overhead and prove the correctness of the RP 3 algorithm. Specifically, we estimate the addional control overhead as 40, 60 and 160 bits per packet in Framework 1, 2 and 3, respectively.…”

“…The RP 3 algorithm under Framework 2 is quite similar to the described RP 3 algorithm under Framework 1, and therefore we detail it in our online technical report [13].…”

“…We now introduce models of the total delay in the RP 3 algorithm under Frameworks 1 and 3 (the detailed proofs are in [13]). To do so, we denote T proc (φ) as the sum of the processing delays of the first φ processing phases of a packet.…”

“…Our online technical report [13] presents additional technical details, including the algorithm under Framework 2, the proofs of correctness of the various algorithms, a discussion of the implementation complexity, and extended simulations.…”

Queueing in the mist: Buffering and scheduling with limited knowledge

“…In our model, similarly to [7], upon processing a U -packet for the first time, its properties become known. We therefore refer to such a first processing cycle of a U -packet as a parsing cycle.…”