Performance comparison of some frameworks.

[gongweibao]

No description provided.

[typhoonzero]

Need to add the unit of this figure

[panyx0718]

I notice the code used for testing grpc is using Paddle's threadpool. Can you change to use pthread similar to the brpc experiment. Also, it would be great to simplify your grpc experiment to be as similar to brpc experiment as possible

[panyx0718]

Could there be extra over head to create threads in threads?
https://github.com/gongweibao/tests/blob/develop/grpc_test/client.cc#L172

[panyx0718]

https://github.com/gongweibao/tests/blob/develop/grpc_test/client.cc#L165
compared with brpc experiment, which creates channel outside timing zone, grpc experiment is creating channel within timing zone

[panyx0718]

https://github.com/Tuvie/brpc/blob/master/benchmark/client.cpp#L49
brpc seems to be using sync call.

https://github.com/gongweibao/tests/blob/develop/grpc_test/client.cc#L71
while grpc is using async with maximum of 3 nested levels of threads.

[gongweibao]

It seems that they are not the key problem of performance.
It open an ISSUE in GRPC some days ago:grpc/grpc#15348

[paddle-bot-old]

Since you haven't replied for a long time, we have closed this issue/pr.
If the problem is not solved or there is a follow-up one, please reopen it at any time and we will continue to follow up.
由于您长期未回复，我们将关闭这个issue/pr。
若问题未解决或有后续问题，请随时重新打开，我们会继续跟进。