On Fri, Sep 18, 2009 at 5:13 PM, Linux Raid Study <linuxraid.study@xxxxxxxxx> wrote: > Hello: > > I'm seeing some strangeness and want some insight. > > Case #1: > I have 5 disks on my system and each is exported to a iSCSI initiator. > When I run IOMeter tool, I see CPU util of about 80% of my CPU for > Seq-Read operations. > > Case #2: > If I bundle these 5 disks as Raid5, the performance is 20% lower as > compared to above case but the CPU util is only 30% for Seq Read > operations (as compared to 80% in case #1) > > Can someone shed some light on this - why would CPU be lower in #2 > (I'm using 2.6.27 kernel in both cases)? You are not comparing equivalent configurations. You are trying to compare the performance of 5-non-striped data disks in case1 to 4 striped data disks in case2. Also, asking for opinions on what might be causing performance differences is almost always the wrong approach. Use a profiler, like oprofile in 2.6.27 or perf in recent kernels, to find the exact answer to where the cpu is spending its time for a given workload. There are also tools like blktrace and seekwatcher that can help analyze block i/o performance. -- Dan -- To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
