I'm glad Aerospike uses this technique. Yes, it is not new. I've been using this method in one form or another (e.g. SIGIO/F_SETOWN) for more than a decade, ho

I'm glad Aerospike uses this technique. Yes, it is not new. I've been using this method in one form or another (e.g. SIGIO/F_SETOWN) for more than a decade, however nginx (relatively modern?) doesn't use this technique and could really benefit from it on some workloads. I wrote about it after I casually mentioned it a roomful of quant programmers and found out they didn't know about it; I do not think most servers use this technique. I've added a note about accept4(). When I have some time I intend to repeat my benchmarks with it to see how it does. Context switching is much cheaper than memory so I do not expect much. You should publish benchmarks that are easy for other people to reproduce. I tried running Aerospike on my laptop and was unable to get above 25,000 writes/sec. I'm certain I'm doing something wrong, because the same hardware easily gets above 60,000 writes/sec on KDB per CPU.