Search

The advent of multicore CPUs has led to renewed interest in software routers built from commodity PC hardware. However fully exploiting the parallelism due to multiple cores requires the ability to efficiently parallelize the delivery of packets to cores. I.e. traffic arriving (departing) on an incoming (outgoing) link at a router is inherently serial and hence we need a mechanism that appropriately demultiplexes (multiplexes) traffic between a serial link and a set of cores. Recent efforts point to modern server network interface cards (NICs) as offering the required mux/demux capability through new hardware classification features. However there has been little evaluation of the extent to which these features match the requirements of software routing from the standpoint of both performance and functionality. This paper takes a first step towards such an evaluation comparing a commodity server NIC to both an idealized "parallel" NIC and a "serial only" NIC. We show that although commodity NICs do improve on serial only NICs they lag an ideal parallel NIC. We find similar gaps in the classification features these NICs offer. We thus conclude with recommendations for NIC modifications that we believe would improve their suitability for software routers.

The advent of multicore CPUs has led to renewed interest in software routers built from commodity PC hardware. However, fully exploiting the parallelism due to multiple cores requires the ability to efficiently parallelize the delivery of packets to cores. I.e., traffic arriving (departing) on an incoming (outgoing) link at a router is inherently serial and hence we need a mechanism that appropriately demultiplexes (multiplexes) traffic between a serial link and a set of cores. Recent efforts point to modern server network interface cards (NICs) as offering the required mux/demux capability through new hardware classification features. However, there has been little evaluation of the extent to which these features match the requirements of software routing from the standpoint of both, performance and functionality. This paper takes a first step towards such an evaluation, comparing a commodity server NIC to both, an idealized "parallel" NIC and a "serial only" NIC. We show that although commodity NICs do improve on serial-only NICs they lag an ideal parallel NIC. We find similar gaps in the classification features these NICs offer. We thus conclude with recommendations for NIC modifications that we believe would improve their suitability for software routers.

Software routers promise to enable the fast deployment of new, sophisticated kinds of packet processing without the need to buy and deploy expensive new equipment. The challenge is offering such programmability while at the same time achieving a competitive level of performance. Recent work has demonstrated that software routers are capable of high performance, but only for conventional, simple workloads (like packet forwarding and IP routing) and, even that, after careful manual calibration. In contrast, we are interested in achieving high performance in the context of a software router running multiple sophisticated packet-processing applications. In particular: first, we identify the main factors that affect packet-processing performance on a modern multicore general-purpose server---cache misses, cache contention, load-balancing across processing cores; then, we formulate an optimization problem that takes as input a particular server architecture and a packet processing flow, and determines how to parallelize the router's functionality across the available cores so as to maximize its throughput.