[noduerme]

I'm working on a web app right now that does a lot of heavy/live/realtime processing with workers. The original thought was to run those workers on the node servers and stream the results over a socket, charging by the CPU/minute. But it surprised me that the performance looks a lot better up to about 16 workers if the user just turns over a few cores to local web workers and runs it on their own box. As long as they don't need 64 cores or something, it's faster to run locally, even on an older machine. Thread transport is slow but sockets are slower at a distance; the bottlenecks are in the main thread anyway. So I've been making sure parts are interchangeable between web workers and "remote socket web workers" assigned to each instance of the app remotely.