📝 SummarXiv

Abstract

With the advent of programmable network hardware, more and more functionality can be moved from software running on general purpose CPUs to the NIC. Early NICs only allowed offloading fixed functions like checksum computation. Recent NICs like the Nvidia Bluefield-3 allow a fully programmable dataplane. In this paper, we present our first steps towards a load balancer named XenoFlow running on the Bluefield-3. Furthermore, we show the capabilities and limitations of the Bluefield-3 eSwitch. Our results show that the Bluefield-3 will not achieve line rate with only 2 entries in a Flow Pipe. However, we also show the adventages of hardware offloading on the NIC and being closer to the network. With XenoFlow, we achieve an 44% lower latency compared to a comparable eBPF-based load balancer running on the host. Furthermore, XenoFlow achieves this low latency even under high load.