On Mon, May 23, 2016 at 10:12 AM, David <dclistslinux@xxxxxxxxx> wrote: > Hi All > > I'm doing some testing with OpenSUSE Leap 42.1, it ships with kernel 4.1.12 > but I've also tested with 4.1.24 > > When I map an image with the kernel RBD client, max_sectors_kb = 127. I'm > unable to increase: > > # echo 4096 > /sys/block/rbd0/queue/max_sectors_kb > -bash: echo: write error: Invalid argument > > I'm seeing very poor sequential read performance which I suspect is a result > of max_sectors_kb being stuck at 127 > > Does anyone know what's going on here? > > The behavior I've seen on older kernels is max_sectors_kb is set to 512 but > can be increased up to max_hw_sectors_kb. On newer kernels max_sectors_kb > matches the image's object size. > > The image is default 4MB object size, here's everything in > /sys/block/rbd0/queue: > > # grep -r . > nomerges:0 > logical_block_size:512 > rq_affinity:1 > discard_zeroes_data:1 > max_segments:128 > max_segment_size:4194304 > rotational:0 > scheduler:none > read_ahead_kb:512 > max_hw_sectors_kb:4096 > discard_granularity:4194304 > discard_max_bytes:4194304 > write_same_max_bytes:0 > max_integrity_segments:0 > max_sectors_kb:127 > physical_block_size:512 > add_random:0 > nr_requests:128 > minimum_io_size:4194304 > hw_sector_size:512 > optimal_io_size:4194304 > iostats:1 This is a bug in the block layer in 4.1 - see [1]. The fix is in 4.2, so upgrading to a later 4.1.* won't help. [1] http://lists.ceph.com/pipermail/ceph-users-ceph.com/2015-December/006878.html Thanks, Ilya _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |