Complete 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!

I 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.

Very cool man. For the real-time updates you keep the browser-client requests open, right? Any reason you did it that way?

12 hour amphetamine binge can produce some good work!