Regarding RPC bench

Hi,

It is interesting to compare these schemes and thanks for the result. I just want to point out the scenario specified here seems not be representative RPC use case -- you initiated a new connection for each rpc. I understood that this was probably because you want to do sorta apple-to-apple comparison between RPC and http1.1. But multiplexing rpcs on a single connection is a very important advantage of rpc framework compared http1.1.

Plus, it would be helpful if you introduce more realistic scenarios such as fixed QPS benchmark.

I saw your guys have pending debate on rpc or rest. The above (multiplexing and fixed QPS benchmark) will show the advantage of rpc.