I'm seeing loopback latency dramatically (>10x) improve after a few seconds of a microbenchmark, dropping from 30-40us to <3us. Is this expected?
I'm running a very simple latency test which sends a single byte between a parent and child process and measures the round-trip latency (using two sockets, one for each direction). At the beginning of the test, the round-trip latency is ~30 us, but this number slowly drops to single-digit values after ~20 seconds. I'm wondering if this is expected behavior, and if so, what it causing it?
I will try to oprofile it tomorrow, but watching 'top' during the run shows 50-60% of the time on SoftIRQ, which my understanding is that its expected. CPU affinity does not seem to make a difference.
I tried TCP and UDP, binding to 127.0.0.1, INADDR_ANY, and a public interface IP, but the behavior was consistent.
This was on a 2-cpu westmere workstation (xeon 5600), running rhel5.3 (2.6.18-128? I think).
thanks,
-Kenny
--
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
