AB is using HTTP/1.0, so it isn't able to do keepalives and it may be the case that you're really benchmarking your TCP stack, instead of your webserver... as TCP connect is probably the bottleneck here :) Not sure how it works under the hood, but enabling tcp_reuse, and tcp_recycle might improve the results for ab, which are irrelevant anyway.
Basically this seem related to HTTP/1.0 being outdated, so it isn't able to handle large amount of connections between 2 points well. We had same characteristics, seen in logs - untill we enabled keepalive and HTTP/1.1 for our micro-services. -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.