Slow gRPC communication with large file in Python

[Josh Parks]

I'm trying to do a large array transfer (10-50MB) over gRPC in python and it's quite slow (5-10 seconds, both client and server on localhost). I've tried both streaming and unary requests, and they both seem to run slowly.

For more details/conversation, here's the stackoveflow question: https://stackoverflow.com/questions/70993553/grpc-slow-serialization-on-large-dataset

And for the minimum reproducible example: https://github.com/parksj10/grpc_min

Any help/guidance much appreciated!!!

[Richard Belleville]

Josh,

I don't think I'm able to reproduce with your repo. I'm getting something like 0.2s on my desktop:

(venv) rbellevi@rbell:~/Dev/tmp/grpc_min$ python3 grpc_client.py
0.28313207626342773s    18506294 photons in 1008640 bins
0.14323067665100098s    18506294 photons in 1008640 bins
(venv) rbellevi@rbell:~/Dev/tmp/grpc_min$ python3 grpc_client.py
0.23985695838928223s    18506294 photons in 1008640 bins
0.13980460166931152s    18506294 photons in 1008640 bins

Also, your requirements.txt includes "grpc=1.0.0". I'm assuming this is just an typo. I used "grpcio".

Maybe try running cProfile to generate a profile of the repro on your machine and sharing that here?

--
You received this message because you are subscribed to the Google Groups "grpc.io" group.
To unsubscribe from this group and stop receiving emails from it, send an email to grpc-io+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/grpc-io/b713e084-a3d9-46e3-aae2-8501164ca449n%40googlegroups.com.

[Lidi Zheng]

Windows tends to have slower "localhost" transport comparing to Linux. Josh, can you share your deployment setup?

The answer in SO is also helpful, streaming calls perform better in terms of throughput.

[Josh Parks]

Hi all, 

I'm on a Macbook M1Max (perhaps it's the architecture causing the issue?), have 64 GB of RAM as well...

Sorry for the error in the requirements,  I used `pyreqs` and that seems to mess up...

Running using pyenv virtual environment

Also, added cprofile outputs for server and client to repo

[Josh Parks]

Indeed it does seem to have to do with the M1 mac I'm developing on... I just ran it on my windows box and got the following:

(ve-pas) C:\Users\Josh\Code\grpc_min>python grpc_client.py                                                              

0.3099939823150635s     18506294 photons in 1008640 bins                                                                

0.2660081386566162s     18506294 photons in 1008640 bins   

Is there a way to fix this? This slow of speed is definitely going to inhibit development on my laptop...