Ceph test cluster, how to estimate performance.
Hi Ceph users, Im working on common lisp client utilizing rados library. Got some results, but don't know how to estimate if i am getting correct performance. I'm running test cluster from laptop - 2 OSDs - VM, RAM 4Gb, 4 vCPU each, monitors and mgr are running from the same VM(s). As for storage, i have Samsung SSD 860 Pro, 512G. Disk is splitted into 2 logical volumes (LVMs), and that volumes are attached to VMs. I know that i can't expect too much from that layout, just want to know if im getting adequate numbers. Im doing read/write operations on very small objects - up to 1kb. In async write im getting ~7.5-8.0 KIOPS. Synchronouse read - pretty much the same 7.5-8.0 KIOPS. Async read is segfaulting don't know why. Disk itself is capable to deliver well above 50 KIOPS. Difference is magnitude. Any info is more welcome. Daniel Mezentsev, founder (+1) 604 313 8592. Soleks Data Group. Shaping the clouds.
Hello Daniel, just throw away your crappy Samsung SSD 860 Pro. It won't work in an acceptable way. See https://docs.google.com/spreadsheets/d/1E9-eXjzsKboiCCX-0u0r5fAjjufLKayaut_F... for a performance indication of individual disks. -- Martin Verges Managing director Mobile: +49 174 9335695 E-Mail: martin.verges@croit.io Chat: https://t.me/MartinVerges croit GmbH, Freseniusstr. 31h, 81247 Munich CEO: Martin Verges - VAT-ID: DE310638492 Com. register: Amtsgericht Munich HRB 231263 Web: https://croit.io YouTube: https://goo.gl/PGE1Bx Am Di., 13. Okt. 2020 um 07:31 Uhr schrieb Daniel Mezentsev <dan@soleks.com
:
Hi Ceph users,
Im working on common lisp client utilizing rados library. Got some results, but don't know how to estimate if i am getting correct performance. I'm running test cluster from laptop - 2 OSDs - VM, RAM 4Gb, 4 vCPU each, monitors and mgr are running from the same VM(s). As for storage, i have Samsung SSD 860 Pro, 512G. Disk is splitted into 2 logical volumes (LVMs), and that volumes are attached to VMs. I know that i can't expect too much from that layout, just want to know if im getting adequate numbers. Im doing read/write operations on very small objects - up to 1kb. In async write im getting ~7.5-8.0 KIOPS. Synchronouse read - pretty much the same 7.5-8.0 KIOPS. Async read is segfaulting don't know why. Disk itself is capable to deliver well above 50 KIOPS. Difference is magnitude. Any info is more welcome. Daniel Mezentsev, founder (+1) 604 313 8592. Soleks Data Group. Shaping the clouds. _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
Hello Daniel, yes Samsung "Pro" SSD series aren't to much "pro", especially when it's about write IOPS. I would tend to say get some Intel S4510 if you can afford it. It you can't you can still try to activate overprovisioning on the SSD, I would trend to say reserve 10-30% of the SSD for wear leveling (writing). First check the number of sectors with hdparm -N /dev/sdX then set a permanent HPA (host protected area) to the disk. The "p" and no space is important..... hdparm -Np${SECTORS} --yes-i-know-what-i-am-xxxxx /dev/sdX Wait a little (!), power cycle and re-check the disk with hdparm -N /dev/sdX. My Samsung 850 Pro are a little reluctant to accept the setting, but after some tries or a little waiting the change gets permanent. At least the Samsung 850 pro stopped to die suddenly with that setting. Without it the SSD occasionally disconnected from the bus and reappeared after power cycle. I suspect it ran of of wear something. HTH, derjohn On 13.10.20 08:41, Martin Verges wrote:
Hello Daniel,
just throw away your crappy Samsung SSD 860 Pro. It won't work in an acceptable way.
See https://docs.google.com/spreadsheets/d/1E9-eXjzsKboiCCX-0u0r5fAjjufLKayaut_F... for a performance indication of individual disks.
-- Martin Verges Managing director
Mobile: +49 174 9335695 E-Mail: martin.verges@croit.io Chat: https://t.me/MartinVerges
croit GmbH, Freseniusstr. 31h, 81247 Munich CEO: Martin Verges - VAT-ID: DE310638492 Com. register: Amtsgericht Munich HRB 231263
Web: https://croit.io YouTube: https://goo.gl/PGE1Bx
Am Di., 13. Okt. 2020 um 07:31 Uhr schrieb Daniel Mezentsev <dan@soleks.com
: Hi Ceph users,
Im working on common lisp client utilizing rados library. Got some results, but don't know how to estimate if i am getting correct performance. I'm running test cluster from laptop - 2 OSDs - VM, RAM 4Gb, 4 vCPU each, monitors and mgr are running from the same VM(s). As for storage, i have Samsung SSD 860 Pro, 512G. Disk is splitted into 2 logical volumes (LVMs), and that volumes are attached to VMs. I know that i can't expect too much from that layout, just want to know if im getting adequate numbers. Im doing read/write operations on very small objects - up to 1kb. In async write im getting ~7.5-8.0 KIOPS. Synchronouse read - pretty much the same 7.5-8.0 KIOPS. Async read is segfaulting don't know why. Disk itself is capable to deliver well above 50 KIOPS. Difference is magnitude. Any info is more welcome. Daniel Mezentsev, founder (+1) 604 313 8592. Soleks Data Group. Shaping the clouds. _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
_______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
-- Andreas John net-lab GmbH | Frankfurter Str. 99 | 63067 Offenbach Geschaeftsfuehrer: Andreas John | AG Offenbach, HRB40832 Tel: +49 69 8570033-1 | Fax: -2 | http://www.net-lab.net Facebook: https://www.facebook.com/netlabdotnet Twitter: https://twitter.com/netlabdotnet
Hi guys, Thanks for replies. I looked through that table - hmmm, that is really true - samsung pro is not really pro. Well im getting what im paying for. Mainly my question was - am i getting adequate to my disk performance, and seems like yes i am. My tests shows 7-8 kIOPs, replicatio factor 2, so it's exactly ~15-16 kIOPs that is strongly correlate with that table. It points me that i don't lose anything on the way to ceph cluster - no issue with client, no issue with network (funny, everything throuhg virtio), and there is no problem with CPU. Only disk is bottle neck. Now i know what exactly is happening with my cluster. Thanks again for help.
Hello Daniel,
yes Samsung "Pro" SSD series aren't to much "pro", especially when it's about write IOPS. I would tend to say get some Intel S4510 if you can afford it. It you can't you can still try to activate overprovisioning on the SSD, I would trend to say reserve 10-30% of the SSD for wear leveling (writing). First check the number of sectors with hdparm -N /dev/sdX then set a permanent HPA (host protected area) to the disk. The "p" and no space is important.....
hdparm -Np${SECTORS} --yes-i-know-what-i-am-xxxxx /dev/sdX
Wait a little (!), power cycle and re-check the disk with hdparm -N /dev/sdX. My Samsung 850 Pro are a little reluctant to accept the setting, but after some tries or a little waiting the change gets permanent.
At least the Samsung 850 pro stopped to die suddenly with that setting. Without it the SSD occasionally disconnected from the bus and reappeared after power cycle. I suspect it ran of of wear something.
HTH,
derjohn
On 13.10.20 08:41, Martin Verges wrote:
Hello Daniel,
just throw away your crappy Samsung SSD 860 Pro. It won't work in an acceptable way.
See https://docs.google.com/spreadsheets/d/1E9-eXjzsKboiCCX-0u0r5fAjjufLKayaut_F... for a performance indication of individual disks.
-- Martin Verges Managing director
Mobile: +49 174 9335695 E-Mail: martin.verges@croit.io Chat: https://t.me/MartinVerges
croit GmbH, Freseniusstr. 31h, 81247 Munich CEO: Martin Verges - VAT-ID: DE310638492 Com. register: Amtsgericht Munich HRB 231263
Web: https://croit.io YouTube: https://goo.gl/PGE1Bx
Am Di., 13. Okt. 2020 um 07:31 Uhr schrieb Daniel Mezentsev <dan@soleks.com
: Hi Ceph users,
Im working on common lisp client utilizing rados library. Got some results, but don't know how to estimate if i am getting correct performance. I'm running test cluster from laptop - 2 OSDs - VM, RAM 4Gb, 4 vCPU each, monitors and mgr are running from the same VM(s). As for storage, i have Samsung SSD 860 Pro, 512G. Disk is splitted into 2 logical volumes (LVMs), and that volumes are attached to VMs. I know that i can't expect too much from that layout, just want to know if im getting adequate numbers. Im doing read/write operations on very small objects - up to 1kb. In async write im getting ~7.5-8.0 KIOPS. Synchronouse read - pretty much the same 7.5-8.0 KIOPS. Async read is segfaulting don't know why. Disk itself is capable to deliver well above 50 KIOPS. Difference is magnitude. Any info is more welcome. Daniel Mezentsev, founder (+1) 604 313 8592. Soleks Data Group. Shaping the clouds. _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
_______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
-- Andreas John net-lab GmbH | Frankfurter Str. 99 | 63067 Offenbach Geschaeftsfuehrer: Andreas John | AG Offenbach, HRB40832 Tel: +49 69 8570033-1 | Fax: -2 | http://www.net-lab.net
Facebook: https://www.facebook.com/netlabdotnet Twitter: https://twitter.com/netlabdotnet
_______________________________________________ ceph-users mailing list -- ceph-users@ceph.ioTo unsubscribe send an email to ceph-users-leave@ceph.io Daniel Mezentsev, founder (+1) 604 313 8592. Soleks Data Group. Shaping the clouds.
participants (3)
-
Andreas John
-
Daniel Mezentsev
-
Martin Verges