PyParallel: added support for detecting system memory high/low states and altering behavior accordingly (i.e. hit high memory, stop accepting new connections until the event clears), refactored the heap snapshot logic, implemented socket re-use and context re-use for socket servers, switched over to using custom threadpools per socket server such that min/max threads could be limited to ncpu (prevents the kernel from flipping out and creating 200-300 threadpool threads when hitting instantaneous load of 10k+ connections (which happened when I was just palming everything off to the default thread pool, which has no min/max thread bounds and simply tries to do "best effort" servicing of thread pool load, which is completely sufficient in just about every case other than huge instantaneous loads)). Removed the extensive pointer/memory address testing from the release build (still in debug build) which, as expected, gave a significant performance improvement. End result, gloriously low latency and low jitter: https://twitter.com/trentnelson/status/562839986408800257. Only crashes now when you ctrl-c it on the console (as I haven't written the cleanup code yet) -- once that is fixed, I'll build an installer and do a public release, wahey! I love it when a plan comes together.

(PyParallel: native CPython running on all cores without being impeded by the GIL. https://speakerdeck.com/trent/pyparallel-how-we-removed-the-...)