HTTP/3 Load Generator

[MDA2AV]

Currently we are using oha, this load generator has some performance limitations and is not able to saturate the frameworks.

If anyone knows a better option please mention here.

[jerrythetruckdriver]

The 38x gap between H2 (3.2M) and H3 (85K) on nginx isn't just the load generator — it's UDP. QUIC needs kernel-level batching to compete: recvmsg multishot, GSO for send coalescing, and ideally io_uring for the syscall amortization that makes UDP viable at scale. Any H3 load generator you pick will bottleneck in the kernel's UDP path before it saturates the framework unless it's using sendmmsg/recvmmsg at minimum.

Check if oha is doing single-datagram sendmsg per QUIC packet — that alone would explain the numbers.