Very simple benchmark of tonic and grpc-rs (aka grpcio).
cargo build --release && cargo run --release --bin test <num-requests>
A single grpc service is implemented, with a single method. This method takes an empty request message and responds with a large message containing 2 MiB of random data. (This emulates a case I happen to care about, where a grpc server is acting in part as a file cache.) The random data is generated in advance so it is the same for all calls to the server; the intent here is to have the server method not take any significant time other than having to send 2 MiB of data.
The test is run four times with all the combinations of grpcio and tonic for client and server.
cargo run --release --bin test 1000
test Grpcio server with Grpcio client
1.797s
test Grpcio server with Tonic client
3.327s
test Tonic server with Grpcio client
7.246s
test Tonic server with Tonic client
6.278s
Perhaps also of interest is that with higher values of
<num-requests>, e.g. 2000, tonic doesn't work.