ceph rbd disk performance question
HI all! I use ceph as the openstack VM disk. I have a VM run postgresql. I found the disk on the vm run postgresql is very busy and slow! But the ceph cluster is very healthy and without any slow request. Even the vm disk is very busy, but the ceph cluster is look like very idle. My ceph version is 12.2.8 and the vm disk is ext4 file system. The postgresql vm disk is very busy! look below: avg-cpu: %user %nice %system %iowait %steal %idle 0.53 0.00 1.6 16.55 0.00 81.31 Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s avgrq-sz avgqu-sz await r_await w_await svctm %util vdb 0.00 7425.00 2.0 65.5 40.00 63904.00 940.54 134.27 66966.54 12553.40 69042.94 14.71 100.05 The ceph cluster is very idle. osd commit_latency(ms) apply_latency(ms) 39 0 1 38 0 2 37 0 0 36 0 1 35 0 0 34 0 0 33 0 0 32 0 0 31 0 0 30 0 1 29 0 1 28 0 0 27 0 1 26 0 1 25 0 1 24 0 1 23 0 0 22 0 0 9 0 0 8 0 0 7 0 0 6 0 1 5 0 1 4 0 7 0 0 1 1 0 3 2 0 2 3 0 1 10 0 1 11 0 0 13 0 0 14 0 0 15 0 1 16 0 0 17 0 1 18 0 0 19 0 0 20 0 0 21 0 0 Can anybody tell me why? Thanks in advance!
You aren't showing the I/O size, only the latency. It looks like this is mostly sequential writes, since it's merging most of the I/O. Because you only assigned 1 volume (sdb), you will be limited to a single queue. I'd recommend adding more volumes and striping the data across. On Sat, Aug 24, 2019, 11:00 AM linghucongsong <linghucongsong@163.com> wrote:
HI all!
I use ceph as the openstack VM disk. I have a VM run postgresql.
I found the disk on the vm run postgresql is very busy and slow!
But the ceph cluster is very healthy and without any slow request.
Even the vm disk is very busy, but the ceph cluster is look like very idle.
My ceph version is 12.2.8 and the vm disk is ext4 file system.
The postgresql vm disk is very busy! look below:
avg-cpu: %user %nice %system %iowait %steal %idle 0.53 0.00 1.6 16.55 0.00 81.31
Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s avgrq-sz avgqu-sz await r_await w_await svctm %util vdb 0.00 7425.00 2.0 65.5 40.00 63904.00 940.54 134.27 66966.54 12553.40 69042.94 14.71 100.05
The ceph cluster is very idle. osd commit_latency(ms) apply_latency(ms) 39 0 1 38 0 2 37 0 0 36 0 1 35 0 0 34 0 0 33 0 0 32 0 0 31 0 0 30 0 1 29 0 1 28 0 0 27 0 1 26 0 1 25 0 1 24 0 1 23 0 0 22 0 0 9 0 0 8 0 0 7 0 0 6 0 1 5 0 1 4 0 7 0 0 1 1 0 3 2 0 2 3 0 1 10 0 1 11 0 0 13 0 0 14 0 0 15 0 1 16 0 0 17 0 1 18 0 0 19 0 0 20 0 0 21 0 0
Can anybody tell me why?
Thanks in advance!
_______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
On 24.08.2019 18:00, Brett Chancellor wrote:
You aren't showing the I/O size, only the latency. It looks like this is mostly sequential writes, since it's merging most of the I/O. Because you only assigned 1 volume (sdb), you will be limited to a single queue. I'd recommend adding more volumes and striping the data across.
Is adding more volumes, and stiping them in OS, better then using the RBD fancy striping options ?
Do both. Host striping will give you more queues, RBD striping will use more OSDs. On Sat, Aug 24, 2019, 12:39 PM Ronny Aasen <ronny+ceph-users@aasen.cx> wrote:
On 24.08.2019 18:00, Brett Chancellor wrote:
You aren't showing the I/O size, only the latency. It looks like this is mostly sequential writes, since it's merging most of the I/O. Because you only assigned 1 volume (sdb), you will be limited to a single queue. I'd recommend adding more volumes and striping the data across.
Is adding more volumes, and stiping them in OS, better then using the RBD fancy striping options ?
_______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
participants (3)
-
Brett Chancellor
-
linghucongsong
-
Ronny Aasen