• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# C++ Performance Notes
2
3## Streaming write buffering
4
5Generally, each write operation (Write(), WritesDone()) implies a syscall.
6gRPC will try to batch together separate write operations from different
7threads, but currently cannot automatically infer batching in a single stream.
8
9If message k+1 in a stream does not rely on responses from message k, it's
10possible to enable write batching by passing a WriteOptions argument to Write
11with the buffer_hint set:
12
13~~~{.cpp}
14stream_writer->Write(message, WriteOptions().set_buffer_hint());
15~~~
16
17The write will be buffered until one of the following is true:
18- the per-stream buffer is filled (controllable with the channel argument
19  GRPC_ARG_HTTP2_WRITE_BUFFER_SIZE) - this prevents infinite buffering leading
20  to OOM
21- a subsequent Write without buffer_hint set is posted
22- the call is finished for writing (WritesDone() called on the client,
23  or Finish() called on an async server stream, or the service handler returns
24  for a sync server stream)
25
26## Completion Queues and Threading in the Async API
27
28Right now, the best performance trade-off is having numcpu's threads and one
29completion queue per thread.
30