Faster iscsi for vmware needed
R3LZX
50 Posts
October 21, 2019, 12:45 pmQuote from R3LZX on October 21, 2019, 12:45 pmok so I have everything running on my three DL380s with single 1TB SSD journal times four 10TB drives on each server, scrubbing took some work to get it sorted out and now its working fine. iops are however naturally a problem.
So what I want to do is purchase another 1TB SSD journal to four 2TB ssd drives per server and have all of it separate for performance. I have looked through the administrator guide on new pool creation but wanted to post here to make sure I do this correctly.
ok so I have everything running on my three DL380s with single 1TB SSD journal times four 10TB drives on each server, scrubbing took some work to get it sorted out and now its working fine. iops are however naturally a problem.
So what I want to do is purchase another 1TB SSD journal to four 2TB ssd drives per server and have all of it separate for performance. I have looked through the administrator guide on new pool creation but wanted to post here to make sure I do this correctly.
admin
2,930 Posts
October 21, 2019, 1:49 pmQuote from admin on October 21, 2019, 1:49 pmIf you use SSD OSDs, you should not use an SSD journal, do not use an external journal for them..unless you use a much faster nvme.
If i understand correctly, you wish to create a separate all SSD pool, you could do this by choosing the by-host-ssd placement rule for the new pool. The issue however is with the existing pool, unless its existing rule was by-host-hdd, the original pool will write to the new SSDs as well if its rule id the default rule which writes on all disks.
Not sure it is related, but it is always best to try to use the same disk model and size as much as possible for best performance.
If you use SSD OSDs, you should not use an SSD journal, do not use an external journal for them..unless you use a much faster nvme.
If i understand correctly, you wish to create a separate all SSD pool, you could do this by choosing the by-host-ssd placement rule for the new pool. The issue however is with the existing pool, unless its existing rule was by-host-hdd, the original pool will write to the new SSDs as well if its rule id the default rule which writes on all disks.
Not sure it is related, but it is always best to try to use the same disk model and size as much as possible for best performance.
R3LZX
50 Posts
October 22, 2019, 9:06 pmQuote from R3LZX on October 22, 2019, 9:06 pmso that pretty much kills that idea, how about adding more journals SSD's so I have 2TB for every four 10TB drives?
so that pretty much kills that idea, how about adding more journals SSD's so I have 2TB for every four 10TB drives?
admin
2,930 Posts
October 22, 2019, 10:39 pmQuote from admin on October 22, 2019, 10:39 pmFor journals, it is the ratio of SSDs to HDDs that count, not so much the size, it is a performance io rate issue.
The 2 pool approach can still be done, first you need to adjust the crush map to tweak your current rule to select hdd devices only, see
https://docs.ceph.com/docs/master/rados/operations/crush-map-edits/#crush-reclassify
For journals, it is the ratio of SSDs to HDDs that count, not so much the size, it is a performance io rate issue.
The 2 pool approach can still be done, first you need to adjust the crush map to tweak your current rule to select hdd devices only, see
https://docs.ceph.com/docs/master/rados/operations/crush-map-edits/#crush-reclassify
Faster iscsi for vmware needed
R3LZX
50 Posts
Quote from R3LZX on October 21, 2019, 12:45 pmok so I have everything running on my three DL380s with single 1TB SSD journal times four 10TB drives on each server, scrubbing took some work to get it sorted out and now its working fine. iops are however naturally a problem.
So what I want to do is purchase another 1TB SSD journal to four 2TB ssd drives per server and have all of it separate for performance. I have looked through the administrator guide on new pool creation but wanted to post here to make sure I do this correctly.
ok so I have everything running on my three DL380s with single 1TB SSD journal times four 10TB drives on each server, scrubbing took some work to get it sorted out and now its working fine. iops are however naturally a problem.
So what I want to do is purchase another 1TB SSD journal to four 2TB ssd drives per server and have all of it separate for performance. I have looked through the administrator guide on new pool creation but wanted to post here to make sure I do this correctly.
admin
2,930 Posts
Quote from admin on October 21, 2019, 1:49 pmIf you use SSD OSDs, you should not use an SSD journal, do not use an external journal for them..unless you use a much faster nvme.
If i understand correctly, you wish to create a separate all SSD pool, you could do this by choosing the by-host-ssd placement rule for the new pool. The issue however is with the existing pool, unless its existing rule was by-host-hdd, the original pool will write to the new SSDs as well if its rule id the default rule which writes on all disks.
Not sure it is related, but it is always best to try to use the same disk model and size as much as possible for best performance.
If you use SSD OSDs, you should not use an SSD journal, do not use an external journal for them..unless you use a much faster nvme.
If i understand correctly, you wish to create a separate all SSD pool, you could do this by choosing the by-host-ssd placement rule for the new pool. The issue however is with the existing pool, unless its existing rule was by-host-hdd, the original pool will write to the new SSDs as well if its rule id the default rule which writes on all disks.
Not sure it is related, but it is always best to try to use the same disk model and size as much as possible for best performance.
R3LZX
50 Posts
Quote from R3LZX on October 22, 2019, 9:06 pmso that pretty much kills that idea, how about adding more journals SSD's so I have 2TB for every four 10TB drives?
so that pretty much kills that idea, how about adding more journals SSD's so I have 2TB for every four 10TB drives?
admin
2,930 Posts
Quote from admin on October 22, 2019, 10:39 pmFor journals, it is the ratio of SSDs to HDDs that count, not so much the size, it is a performance io rate issue.
The 2 pool approach can still be done, first you need to adjust the crush map to tweak your current rule to select hdd devices only, see
https://docs.ceph.com/docs/master/rados/operations/crush-map-edits/#crush-reclassify
For journals, it is the ratio of SSDs to HDDs that count, not so much the size, it is a performance io rate issue.
The 2 pool approach can still be done, first you need to adjust the crush map to tweak your current rule to select hdd devices only, see
https://docs.ceph.com/docs/master/rados/operations/crush-map-edits/#crush-reclassify