7 ms·
12 hours in and the site is up: https://josephg.com/sp/ https://josephg.com/sp/ Polish and perf tuning tomorrow. Code: https://github.com/josephg/sephsplace h
by josephg 10y ago
12 hours in and the site is up: https://josephg.com/sp/ https://josephg.com/sp/
Polish and perf tuning tomorrow.
Code: https://github.com/josephg/sephsplace https://github.com/josephg/sephsplace (warning: contains evil)
- always_good 10y ago12 hour amphetamine binge can produce some good work!
- renlo 10y agoVery cool man. For the real-time updates you keep the browser-client requests open, right? Any reason you did it that way?
- josephg 10y agoWell, the server needs to message the client with updates somehow. Either the client keeps a connection open or it has to periodically poll the client. Keeping a connection open is faster and it uses less system resources.
- josephg 10y agoComplete and working. I haven't tested the load for real, but judging by the performance numbers it should be able to handle 100k connections fine with 2 beefy backend servers. (40% of one of my laptop's cores handles 15k connections with 400 edits per second. So about 8 cores should handle 100k users no sweat.) The bottleneck is fanout, so until kafka starts hitting subscribe limits it should scale linearly with frontend servers. (And kafka is a long way from hitting that limit because I've added write coalescing.) I'll post a writeup about it all once I've eaten some real food and had a rest. But for now, enjoy!
- socmag 10y agoI like what you are doing a lot. It doesn't matter that it isn't a one to one comparison, the point is you are taking on the challenge and enjoying it. Probably actually you are learning some stuff along the way as well. It would be kind of neat if the place experiment could become something of a micro-benchmark for online customer facing distributed platforms. Keep up the good work man, it's looking good. Don't forget to hydrate and consume pizza.