On Mon, 30 Apr 2007, Quan Sun wrote: > Hi, > On 4/30/07, pradeep singh <2500.pradeep@xxxxxxxxx> wrote: > > heh... if a lot of disk I/O is required try changing the I/O scheduler too. > > Make it CFQ, will help i guess.Read about Anticipatory and other IO > > schedulers too. Though IMHO CFQ will be good enough. > > The problem is, my application is a heavy network socket IO proxy, and > no disk IO involved. :) Cat /proc/interrupts to see what interrupt the network interface is using, and if interrupts are being load balanced across CPUs. Then cat /proc/irq/XX/smp_affinity where XX is the interrupt number determined from the cat /proc/interrupts. You can explicitly set this also, but I'm not sure you would want to. For the application you can use sched_setaffinity(). In addition to the noirqbalance kernel parameter, you may want to disable the irqbalance daemon. On a RedHat based system you would do this via: chkconfig --level 0123456 irqbalance off /etc/init.d/irqbalance stop You may want to test both with and without the irqbalance daemon running, to see which gives better overall performance. However, when running the irqbalance daemon, it will override any explicit SMP affinity settings you might have made. -Bill - To unsubscribe from this list: send the line "unsubscribe linux-net" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
