On Thu, 2008-08-28 at 12:29 +0530, Sukanto Ghosh wrote: > Hi, > > What was the reason for changing the CFS mechanism from > fair_clock/wait_runtime to vruntime? Numerical integrety mostly. > What were the kind of workloads where the fairclock approach faired badly ? They are analytically identical - its the discrete version of the fair_clock code that led to unfixable numerical pain. > Also, when adding a blocked task to the runqueue (rbtree), why is it's > vruntime set to less than min_vruntime ? > This means it would preempt the currently running process. > If we have some processes which run for some (very less) time and get > blocked repeatedly, won't the processes at the end of runqueue starve > .... because some or the other blocked process gets unblocked and gets > the cpu (as vruntime is lesser than min_vruntime). It never gets puts left of its previous value, so starvation cannot happen. -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
