Benchmarking

pw_rpc: Efficient, low-code-size RPC system for embedded devices

pw_rpc provides an RPC service and Python utilities for measuring throughput, latency, and resilience of an RPC deployment over its underlying transport.

pw.rpc.Benchmark service

The Benchmark service provides a low-level RPC interface for sending bulk data between client and server. The service is defined in pw_rpc/benchmark.proto.

A raw RPC implementation of the benchmark service is provided, suitable for any system running pw_rpc.

To use the benchmark service in C++:

#include "pw_rpc/benchmark.h"
#include "pw_rpc/server.h"

constexpr pw::rpc::Channel kChannels[] = {/* ... */};
static pw::rpc::Server server(kChannels);

static pw::rpc::BenchmarkService benchmark_service;

void RegisterServices() { server.RegisterService(benchmark_service); }

Benchmark service definition

syntax = "proto3";

package pw.rpc;

service Benchmark {
  // The server responds with the payload the client sent.
  rpc UnaryEcho(Payload) returns (Payload);

  // The server responds to each request payload the client sends. The client
  // stops the RPC by cancelling it.
  rpc BidirectionalEcho(stream Payload) returns (stream Payload);
}

message Payload {
  bytes payload = 1;
}

Stress testing

The Benchmark service is also used to fuzz and stress-test the pw_rpc module across multiple concurrent threads and fluctuating channel conditions.

To run the client-server fuzz stress test:

bazelisk test //pw_rpc/fuzz:cpp_client_server_fuzz_test