Benchmarked on a single 6 core / 12 thread machine, with 9 client threads and 1-3 server threads. Language wars aside, it's hard to extract useful information l

Benchmarked on a single 6 core / 12 thread machine, with 9 client threads and 1-3 server threads. Language wars aside, it's hard to extract useful information like CPU time overhead on the server from these results. For the single core test, we can infer an upper bound of 20-33us of server cpu time per request for most of the relevant languages for gRPC. That seems pretty good.