Apparently this piece of amazing Google technology is actually being publicized. Maybe this presentation doesn't describe its importance or its applications well enough (200 views? seriously??), but the basic idea is: it's lightweight native threads, with all of the benefits but zero of the typical shortcomings.
You get 200-nanosecond cost for creating or switching threads (i.e. it costs as much to create a thread or switch contexts than to add a callback to a thread pool - that's so cheap you can have not just a thread for every user request, but a thread for every request you make to a backend, i.e. you can forget about asynchronous programming style forever
); you get userlevel-controllable scheduling; you get core locality for related threads and better latency; AND this all completely transparently to all other things that deal with threads: the OS's native preemptive scheduler still works; you can do arbitrary blocking stuff; all debuggers, thread-local variables, stack traces etc. just transparently continue to work without modification.
I think this is insanely cool.User-level threads....... with threads. - Paul Turner - Google