Context switch and kernel scheduler implemention details also dominate the networking busy loop tests with one client. The kind of naive benchmark you see often

Context switch and kernel scheduler implemention details also dominate the networking busy loop tests with one client. The kind of naive benchmark you see often in blog posts where every database tested seems to deliver like 10k ops per sec. You are actually just measuring the latency in the networking ping pong the kernel is able to sustain between two processes, so usually the fastest client implementation wins (because the rest is the same for all the systems, being outside the systems).