Publication details

N. Mulè, J. Carlon, G. Lettieri, T. Cucinotta. "A Kernel-Bypass Data Path for the DistWalk Distributed Workload Emulator," in Proceedings of the 21st Workshop on Virtualization in High-Performance Cloud Computing (VHPC 2026), 32nd International European Conference on Parallel and Distributed Processing (Euro-Par 2026), August 24th, 2026, Pisa, Italy.

Abstract

With the growth of distributed systems in size and complexity, evaluating their end-to-end performance has become increasingly challenging. DistWalk is an open-source distributed workload emulator allowing for estimating the impact of key design decisions on the performance of an application, prior to its development. For performance-critical software, the overhead introduced by the networking stack has led to a broad adoption of kernel-bypass techniques. To enable DistWalk to model such deployments, we integrated a data path based on the Data Plane Development Kit (DPDK), a kernel-bypass framework. The DPDK code path is fully compatible with the existing socket-based implementation, letting both coexist within the same binary and allowing cross-protocol forwarding between DPDK and TCP/UDP nodes in the same topology. Support for multi-threading is provided by leveraging Receive Side Scaling (RSS) to dispatch incoming packets to different worker threads according to their source MAC address. On two servers connected through a 10 Gbps Ethernet link, DPDK achieved a median round-trip latency of 11.2 µs, approximately 2.7 times faster than UDP and 3 times faster than TCP in their best achievable configurations. The new feature is available in the original DistWalk repository.

Copyright by IEEE.

Download paper