This package contains naive implementation of benchmarks for tonic

Test with ghz

cargo run --bin=server --release

Rust version will start on port 50051

ghz  --insecure --proto proto/hellobench.proto --call hellobench.Greeter.SayEmpty -d '{}' -n 100000 -c 16 --connections=4 localhost:50051

Summary:
  Count:        100000
  Total:        6.26 s
  Slowest:      100.07 ms
  Fastest:      0.16 ms
  Average:      0.96 ms
  Requests/sec: 15984.89

Compile and run go version

cd go
go build server.go
server

Go version will start on port 50050, so above ghz should be amended:

ghz  --insecure --proto proto/hellobench.proto --call hellobench.Greeter.SayEmpty -d '{}' -n 100000 -c 16 --connections=4 localhost:50051

Summary:
  Count:        100000
  Total:        4.57 s
  Slowest:      14.45 ms
  Fastest:      0.11 ms
  Average:      0.67 ms
  Requests/sec: 21890.10

Test with Rust client

Load with empty request/response

target/release/client --connections=4 --concurency=8 --request=Empty --messages=100000 --port=50051

Elapsed: 20022ms
processed 400000 with 19978 rps
successful 400000 failed 0 requests

OR Load with string request/response

target/release/client --connections=4 --concurency=8 --request=Something --messages=100000 --port=50051

Elapsed: 22059ms
processed 400000 with 18133 rps
successful 400000 failed 0 requests

Run with flamegraph

sudo flamegraph target/release/server

Comparison with plain text hyper http

Simple naive hyper HTTP server on same machine gives about ~3x more requests/responses

cargo run --release --bin=http_server

❯ ab -n 100000 -c 4 -k http://127.0.0.1:3000/
...
Concurrency Level:      4
Time taken for tests:   3.125 seconds
Requests per second:    32004.64 [#/sec] (mean)
Transfer rate:          3500.51 [Kbytes/sec] received