I wrote my own perf tool and websocket library. Here are some thing I have observed:
* my websocket library achieves performance about 37x faster than bun, but that comparison is not precise. Bun claims to send messages at 165k/sec in August and 159k/sec in Sept. I don’t know their test hardware details or message size and so on, so I cannot say that this is a fair comparison.
* The article is reporting single socket round trip speeds around 50k/sec. If my measurements were a fair comparison my own WebSockets library is 100x faster. Before I learned the secret sauce of transmission speed my WebSockets library was about 9-11x faster at sending than receiving. Now, that gap has disappeared. In my own hardware I can send as fast as 6m/sec on Node and roundtrip it’s 5m/sec. Both the increase in total performance, the reduction of round trip gap, and instability in Node come from the same single observation.
* my WebSocket library executes about 40% faster on node than bun. I cannot execute it on Deno because Deno has some sort of conflict with OpenSSL run as a child process.
* bun is super stable. When I push my WebSocket library up to 6m/sec single socket send speed the experiment frequently breaks node due to memory exhaustion in V8. I have been able to run an experiment of 10 tests each comprising 50m messages on bun. It took about 76min to complete in my hardware but it did complete. I have only gotten bun to crash once.
* I have not tried to run my code on the zig version of bun.
I wrote my own perf tool and websocket library. Here are some thing I have observed:
* my websocket library achieves performance about 37x faster than bun, but that comparison is not precise. Bun claims to send messages at 165k/sec in August and 159k/sec in Sept. I don’t know their test hardware details or message size and so on, so I cannot say that this is a fair comparison.
* The article is reporting single socket round trip speeds around 50k/sec. If my measurements were a fair comparison my own WebSockets library is 100x faster. Before I learned the secret sauce of transmission speed my WebSockets library was about 9-11x faster at sending than receiving. Now, that gap has disappeared. In my own hardware I can send as fast as 6m/sec on Node and roundtrip it’s 5m/sec. Both the increase in total performance, the reduction of round trip gap, and instability in Node come from the same single observation.
* my WebSocket library executes about 40% faster on node than bun. I cannot execute it on Deno because Deno has some sort of conflict with OpenSSL run as a child process.
* bun is super stable. When I push my WebSocket library up to 6m/sec single socket send speed the experiment frequently breaks node due to memory exhaustion in V8. I have been able to run an experiment of 10 tests each comprising 50m messages on bun. It took about 76min to complete in my hardware but it did complete. I have only gotten bun to crash once.
* I have not tried to run my code on the zig version of bun.