rbd create liuliangtest -p vms -size 20G
rbd map -p vms liuliangtest
mkfs.ext4 /dev/rbd0 && mount /dev/rbd0 /mnt
mount -l  show:  /dev/rbd0 on /mnt type ext4 
(rw,relatime,stripe=1024,data=ordered)

test one of ext4: use psync
command: fio --bs=4k --ioengine=psync --direct=1 --rw=randwrite --runtime=60 
--group_reporting --name=fio-write --size=1G --filename=/mnt/test.db 
-numjobs=32 --iodepth=1
result iops: 300

test two of ext4: use libaio
command: fio --bs=4k --ioengine=libaio --direct=1 --rw=randwrite --runtime=60 
--group_reporting --name=fio-write --size=1G --filename=/mnt/test.db 
-numjobs=32 --iodepth=1
result iops: 9000

than I umount /mnt  and mkfs.xfs /dev/rbd0 && mount /dev/rbd0 /mnt
mount -l show: /dev/rbd0 on /mnt type xfs 
(rw,relatime,attr2,inode64,sunit=8192,swidth=8192,noquota)


test one of xfs: use psync
command: fio --bs=4k --ioengine=psync --direct=1 --rw=randwrite --runtime=60 
--group_reporting --name=fio-write --size=1G --filename=/mnt/test.db 
-numjobs=32 --iodepth=1
result iops: 9000

test one of xfs: use libaio
command: fio --bs=4k --ioengine=libaio --direct=1 --rw=randwrite --runtime=60 
--group_reporting --name=fio-write --size=1G --filename=/mnt/test.db 
-numjobs=32 --iodepth=1
result iops: 10000

As the describe above .  When use  libaio  .The iops of ext4 is approximately 
equal to xfs.  But when I use psync . xfs  is 30 times better than ext4.  Is 
this correct ? Who can help explain.
_______________________________________________
ceph-users mailing list
ceph-users@lists.ceph.com
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

Reply via email to