2009/6/6 André Braga <meianoite@xxxxxxxxx>: > scheduler and post the results? Beware that the affine scheduler > sometimes tend to make a terrible job choosing the CPUs to run a > thread on, and sometimes the system lags a lot, but sometimes it works Yes, it's currently suboptimal. I'd intended to work on this further but due to financial issues I had to hold off on my plans for getting a multicore box, which makes testing a bit difficult. Right now the affinity's not quite as hard as it should be, and sometimes there's an anomaly where a task migrates over onto an already loaded CPU and stays there. Without an SMP system to test with though, tuning/correcting those issues is decidedly non-trivial. If someone wants to jump in and has the requisite hardware, they're more than welcome to. By and large I'd be somewhat surprised if the results differed greatly from scheduler_simple in this case though, I suspect there's probably lock contention issues outside of the scheduler somewhere that are causing most of the problems. Regards, Rene