Simple websocket performance tool in Python

Building

./build.sh

Running

First start up the server

./run_server.sh --random-message --compression GZIP --batch-size 100

then a corresponding client

./run_client.sh --compression GZIP --batch-size 100

batch-size and compression need to match between client and server.

Example results

Tested on i5-8250U CPU (4 cores, 1.60GHZ) with 1000000 messages.

Random message	Compression	Batch size	Result
true	GZIP	100	120,000 msg/s
false	GZIP	100	160,000 msg/s
true	None	100	215,000 msg/s
false	None	100	330,000 msg/s
true/false	GZIP	1	14,200 msg/s
true/false	None	1	42,000 msg/s

Batching and no compression offers the best throughput when testing locally. When network connection becomes an issue I would suspect compression would lead to better results. Additionally, a faster compression algorithm could be chosen, as GZIP is known to be slow, however Python lacks proper built-in fast access to algorithms like ZSTD.

Caveats

Given that the functions use only await websocket.send() multiple clients will not be able to stream data at the same time, unless the write buffer is full. This is done deliberately to test the throughput without additional throttling. See: https://websockets.readthedocs.io/en/stable/api.html#websockets.protocol.WebSocketCommonProtocol.send

If you want to see the performance with multiple clients connected, just add await asyncio.sleep(0) to the while True: loop. See: python/asyncio#284 (comment)