PacketMill: toward per-Core 100-Gbps networking

Abstract

We present PacketMill, a system for optimizing software packet processing, which (i) introduces a new model to efficiently manage packet metadata and (ii) employs code-optimization techniques to better utilize commodity hardware. PacketMill grinds the whole packet processing stack, from the high-level network function configuration file to the low-level userspace network (specifically DPDK) drivers, to mitigate inefficiencies and produce a customized binary for a given network function. Our evaluation results show that PacketMill increases throughput (up to 36.4 Gbps ś 70%) & reduces latency (up to 101 µs ś 28%) and enables nontrivial packet processing (e.g., router) at ≈100 Gbps, when new packets arrive > 10× faster than main memory access times, while using only one processing core.

Dynamic Recompilation of Software Network Services with Morpheus

“…Applying Morpheus to a DPDK program, we increase performance by up to 469%. Finally, we measured Morpheus against state-of-the-art network code optimization frameworks such as ESwitch [62] and PacketMill [30]: we show that Morpheus boosts the throughput by up to 80% and 294%, respectively, compared to existing work. Contributions.…”

“…2 The need for tracking packet-level dynamics. The potential to optimize code for specific network configurations has been explored in prior work, for OpenFlow [62], P4 software [75,86] and hardware targets [5], network functions [69], and programmable switches [30] (see Table 1). In order to maximize performance, however, we need to go beyond specializing the code for relatively stable runtime configuration and apply optimizations at the packet level.…”

“…We configured the router with either 20 or 500 rules taken from the Stanford routing tables [49] and generated traffic with different levels of locality as before. We tested the throughput and the latency of the baseline code and the Morpheus optimized one and we compared the results to a state-of-the-art DPDK packet-processing optimizer, PacketMill [30]. In our tests PacketMill uses the following optimizations: removing virtual function calls, inlining variables, and allocating/defining the elements' objects in the source code.…”

“…P2GO [86] Offline profile-guided optimizer for P4/RMT packet-processing pipelines. PacketMill [30] Packet metadata management optimizer for DPDK-based software data planes. NFReducer [69] Policy-driven optimizer for network function virtualization.…”

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

“…Neugebauer et al [33] have investigated the performance of the PCIe device interconnecting modern NICs to the CPUs and memories showing surprisingly low performance with small packets. Farshin et al (i) quantified the impact of direct cache access in Intel processors [9] and (ii) proposed software stack optimizations to achieve per-core hundred-gigabit networking [8]. Kuzniar et al [23,24] have unveiled a variety of issues with the initial OpenFlow-based switches, such as the lack of consistency during updates.…”