On Mon, Oct 12, 2015 at 09:41:18PM +0530, Kashyap Desai wrote: ><various bits snipped> > > In our case, <ignore> and <subset> policy does not work because > <irqbalancer> is designed to consider NUMA node locality. I read below > article and Neil Horman explained default policy of <irqbalancer> will be > moved to <ignore> > http://sourceforge.net/p/e1000/bugs/394/ > > <mpt3sas> driver follows same logic as ixgbe driver. We create multiple > <msix> vector depending upon logical CPU and assign one <msix vector> to > single <logical cpu>. We really do not wants those assignment to be Numa > Node biased. > What should be the solution if we really want to slow down IO submission > to avoid CPU lockup. We don't want only one CPU to keep busy for > completion. > > Any suggestion ? > Yup, file a bug with Oracle :) What you're seeing looks like at least in part a bug with your (very old) version of irqbalance. I seem to recall fixing more than a few bugs dealing with affinity masks from the hint files and banned_cpu options. I strongly suggest that you test with an upstream version of irqbalance and contact oracle to update their version to something more recent. Regards Neil > ` Kashyap > > _______________________________________________ > irqbalance mailing list > irqbalance at lists.infradead.org > http://lists.infradead.org/mailman/listinfo/irqbalance >
