Comparing the performance of state-of-the-art software switches for NFV

Abstract

With the ultimate goal of replacing proprietary hardware appliances with Virtual Network Functions (VNFs) implemented in software, Network Function Virtualization (NFV) has been gaining popularity in the past few years. Software switches route traffic between VNFs and physical Network Interface Cards (NICs). It is of paramount importance to compare the performance of different switch designs and architectures. In this paper, we propose a methodology to compare fairly and comprehensively the performance of software switches. We first explore the design spaces of seven state-of-the-art software switches and then compare their performance under four representative test scenarios. Each scenario corresponds to a specific case of routing NFV traffic between NICs and/or VNFs. In our experiments, we evaluate the throughput and latency between VNFs in two of the most popular virtualization environments, namely virtual machines (VMs) and containers. Our experimental results show that no single software switch prevails in all scenarios. It is, therefore, crucial to choose the most suitable solution for the given use case. At the same time, the presented results and analysis provide a deeper insight into the design tradeoffs and identifies potential performance bottlenecks that could inspire new designs.

Metronome

“…For evaluating the system we used the testbed depicted in Section 3.3, running the l3fwd DPDK application [39] on an isolated NUMA node and generating traffic with MoonGen [40]. For benchmarking our system, we used the evaluation suite provided by Zhang et al in [41]. Tests are done with 64B packets, as this is the worst case scenario 6 .…”

What You Need to Know About (Smart) Network Interface Cards

“…Each core has 2×32 KiB L1 (instruction and data caches) and a 1 MiB L2 cache, while one 25 MiB Last Level Cache (LLC) is shared among the cores in each socket. Following today's best practices, hyper-threading is disabled on all servers [47] and the Operating System (OS) is the Ubuntu 18.04.5 distribution with Linux kernel v4.15. One server acts as a traffic generator and receiver while the other server is the Device Under Test (DUT).…”

PacketMill: toward per-Core 100-Gbps networking

“…A fair comparison between PacketMill and state-of-the-art packet processing frameworks (e.g., BESS [36,37] & VPP [2,27]) requires (i) developing a high-performance NF in pure DPDK, (ii) modifying those other frameworks to enable the case for source-to-source optimizations & LLVM pass that reorders metadata, and (ii) devising scenarios/experiments to avoid any incorrect conclusions, which is beyond the scope of this paper (as done by [102] previously in a separate research publication). However, we have performed a simple comparison to, specifically, show the full potential of X-Change.…”