Benchmarks are invalid

[anzdaddy]

go test doesn't run main(), so you're actually benchmarking the failure case, and for whatever reason, gRPC fails much faster than REST. I've updated the code and added support for over-the-wire benchmarking at https://github.com/anzdaddy/grpc_v_rest.

[husobee]

Oh wow, that is a very different result! Thanks so much for taking the time to look into these tests. I would be curious to know if there is any difference between less trivial payloads. Thanks!

[anzdaddy]

It's not all bad news for gRPC, btw. I've updated my version to cover parallelised tests and a streaming gRPC call. The results were … surprising:

1. Parallelism sped up gRPC 7x, REST only 3x.
2. Sequential GRPC streaming was a whopping 960x faster than sequential request/response.
3. Parallelised streaming only managed a further 1.4x speed-up.

$ GRPC_REMOTE_ADDR=192.168.87.42:4443 REST_REMOTE_ADDR=192.168.87.42:4444 go test -bench Remote -benchtime 10s
goos: darwin
goarch: amd64
pkg: github.com/anzdaddy/grpc_v_rest
BenchmarkGRPCSetInfoRemote-4                    3000       3562536 ns/op
BenchmarkGRPCSetInfoRemote16x-4                50000        287719 ns/op
BenchmarkGRPCSetInfoStreamRemote-4           5000000          3710 ns/op
BenchmarkGRPCSetInfoStreamRemote16x-4        5000000          2725 ns/op
BenchmarkRESTSetInfoRemote-4                    5000       3441740 ns/op
BenchmarkRESTSetInfoRemote16x-4                50000        333709 ns/op
PASS
ok      github.com/anzdaddy/grpc_v_rest    106.145s