Benchmarking Non‐Port‐Hashing Routers With Multiple IP Addresses: Test Setup Recommendations

ABSTRACT

Benchmarking routers with multiple IP addresses is essential for unbiased performance evaluation of devices using Receive‐Side Scaling (RSS) implementations that hash only IP addresses, not port numbers. However, the optimal test configuration for such benchmarking has not yet been established. This paper investigates two test setups: direct connection (where the Tester and the Device Under Test share the same IP network) and gateway‐based connection (where packets are routed through a gateway). Extensive experiments were conducted using IPv4 and IPv6, with Linux and OpenBSD operating systems, and with 16 and 36 CPU cores. The results reveal that the direct connection setup suffers from significant throughput degradation when the number of IP addresses exceeds certain thresholds, likely due to ARP/NDP table scalability limitations. In contrast, the gateway‐based setup consistently achieves stable throughput regardless of the number of IP addresses used. Based on these findings, the authors recommend the gateway‐based test setup with the maximum available IP addresses (approximately 64 k per side) for reliable router benchmarking. This work provides practical guidance for the ongoing IETF BMWG effort to standardize multi‐address benchmarking methodology.