configure number of used hardware threads for synchronous gRPC C++ server

[Christina Müller]

Hi @all,

I have a gRPC application with the following setup:

- bidirectional streaming

- synchronous gRPC server in C++

- gRPC client in go

The server is running in an environment with 2 CPUs:

I have run my application once with NUM_CQS equals to 1 and the second time with NUM_CQS equals 2:

  builder.SetSyncServerOption(ServerBuilder::SyncServerOption::NUM_CQS, 2);

  builder.SetSyncServerOption(ServerBuilder::SyncServerOption::MIN_POLLERS, 1);

  builder.SetSyncServerOption(ServerBuilder::SyncServerOption::MAX_POLLERS, 1);

    builder.SetSyncServerOption(ServerBuilder::SyncServerOption::CQ_TIMEOUT_MSEC, 1000000000);

For the same number of clients making gRPC requests in parallel, there is no difference in throughput for the two runs. Although I would expect that with setting 2, we would have higher throughput.

Is my expectation wrong? How can I control how many cores are used by the synchronous gRPC server?

Thank for inputs!