Christoph Hellwig <[email protected]> writes: > Hi Jeff, > > thanks for the detailed numbers! > > The bigger I/O size makes a drastic impact for Linux software RAID > setups, for which this was a driver. For the RAID5/6 over SATA disks > setups that I was benchmarking this it gives between 20 and 40% better > sequential read and write numbers.
Hi, Christoph, Unfortunately, I'm unable to reproduce your results (though my test setup uses SAS disks, not SATA). I tried with a 10 data disk md RAID5, with 32k and 128k chunk sizes. I modified the fio program to read/write multiples of the stripe width, and I also used aio-stress over a range of queue depth and I/O sizes for read, write, random read and random write. I didn't see any measurable performance difference. Do you still have access to your test setup? What do you think about reinstituting the artificial max_sectors_kb cap, but bumping the default up from 512KB to 1280KB? I had our performance team run numbers on their test setup (which is a 12 disk raid0 with 32k chunk size, fwiw) with max_sectors_kb set to 1280 and, aside from one odd data point, things looked good. Cheers, Jeff -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
