Large Array Serialization using gRPC

Prerequisites

Install gRPC for C++ using by following the directions at gRPC Quick Start. This will install both gRPC and protobuf, which you will need to compile the server and client. If you wish you use the Python server as well, install Python 3.6 or newer.

You will also need cmake to build the server and client, and the directions for installing a modern version of cmake are also at gRPC Quick Start.

C++ Server

Enter the cpp directory and run the following to build the demo_server

mkdir build
cd build
cmake ..
make -j

Next, start the demo_server with:

./demo_server

You should see:

Server listening on 0.0.0.0:50000

The C++ server is now waiting for connections from port 50000.

C++ Client

After compiling the C++ server, start the C++ client with

./demo_client

This will automatically run the streaming demo:

./demo_client
Connected to server at: localhost:50000
Created an INT32 array on the server size 10000000
Array is 38.15 MB

Testing with byte stream...
Using chunk size 256 kB
Average time: 0.0167268
Approx speed: 2.23 GBps

Testing with repeated messages...
Average time: 0.0932537
Approx speed: 409.07 MBps

Testing with repeated chunked messages...
Average time: 0.0300367
Approx speed: 1.24 GBps

Use the -h flag to see the options available when running the client.

./demo_client -h

demo_client [options]
Options:
-h               Print this help
--target         Set the target channel (default localhost:50000)
--chunk_size     Set the chunk size in kB (default 256)
--skip_repeated  Skip testing the repeated messages
--array_size     Size of array in int32
--ntimes_stream  Number of times to test the stream

Python Client

First, ensure you have the following packages installed:

• grpcio
• google-api-python-client

Also, make sure that the version of the packages is greater than or equal to your gRPC C++ version.

Next, enter the python directory and run:

make

This will generate the python gRPC interface files chunkdemo_pb2.py and chunkdemo_pb2_grpc.py. While you have the C++ server running, run:

python client.py

This will connect to the server at your local host and run a basic array transfer test using both a byte stream and repeated messages.

Connected to server at 127.0.0.1:50000
Created an INT32 array on the server size 20000000
Testing with byte stream...
Average time: 0.03763810700038448
Aprox speed: 2.0GiB

Testing with repeated messages...
Average time: 0.8858063033355089
Aprox speed: 86.1MiB