Hi, Does someone know if the following harddisk has a decent performance in a ceph cluster: Micron 5210 ION 1.92TB, SATA (MTFDDAK1T9QDE-2AV1ZABYY) The spec state, that the disk has power loss protection, however, I'd nevertheless like to make sure that all goes well with this disk. Best Regards, Hermann -- hermann@qwer.tk PGP/GPG: 299893C7 (on keyservers)
That depends on how you define “decent” , and your use case. Be careful that these are QLC drives. QLC is pretty new and longevity would seem to vary quite a bit based on op mix. These might be fine for read-mostly workloads, but high-turnover databases might burn them up fast, especially as they fill up.
On Mar 5, 2020, at 12:38 PM, Hermann Himmelbauer <hermann@qwer.tk> wrote:
Hi, Does someone know if the following harddisk has a decent performance in a ceph cluster:
Micron 5210 ION 1.92TB, SATA (MTFDDAK1T9QDE-2AV1ZABYY)
The spec state, that the disk has power loss protection, however, I'd nevertheless like to make sure that all goes well with this disk.
Best Regards, Hermann
-- hermann@qwer.tk PGP/GPG: 299893C7 (on keyservers) _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
I have just ordered two of them to try. (the 3.47GB ION's) If you want, next week I could perhaps run some commands on them..? MJ On 3/5/20 9:38 PM, Hermann Himmelbauer wrote:
Hi, Does someone know if the following harddisk has a decent performance in a ceph cluster:
Micron 5210 ION 1.92TB, SATA (MTFDDAK1T9QDE-2AV1ZABYY)
The spec state, that the disk has power loss protection, however, I'd nevertheless like to make sure that all goes well with this disk.
Best Regards, Hermann
Last monday I performed a quick test with those two disks already, probably not that relevant, but posting it anyway: I created a two-disk ceph 'cluster' on just the one local node, and ran the following:
root@ceph:~# rados bench -p scbench 10 write --no-cleanup hints = 1 Maintaining 16 concurrent writes of 4194304 bytes to objects of size 4194304 for up to 10 seconds or 0 objects Object prefix: benchmark_data_ceph_48906 sec Cur ops started finished avg MB/s cur MB/s last lat(s) avg lat(s) 0 0 0 0 0 0 - 0 1 16 107 91 363.961 364 0.0546517 0.155121 2 16 206 190 379.948 396 0.1529 0.159227 3 16 324 308 410.614 472 0.0972163 0.151421 4 16 458 442 441.942 536 0.0484349 0.141799 5 16 590 574 459.141 528 0.0445051 0.136922 6 16 727 711 473.941 548 0.181066 0.134468 7 16 856 840 479.941 516 0.187683 0.133199 8 16 970 954 476.942 456 0.070753 0.132642 9 16 1089 1073 476.831 476 0.193608 0.133754 10 16 1214 1198 479.142 500 0.0999212 0.132529 Total time run: 10.097218 Total writes made: 1215 Write size: 4194304 Object size: 4194304 Bandwidth (MB/sec): 481.321 Stddev Bandwidth: 60.481 Max bandwidth (MB/sec): 548 Min bandwidth (MB/sec): 364 Average IOPS: 120 Stddev IOPS: 15 Max IOPS: 137 Min IOPS: 91 Average Latency(s): 0.132889 Stddev Latency(s): 0.0645579 Max latency(s): 0.336118 Min latency(s): 0.0117049
Do let me know what else you'd want me to do. MJ
Hi, Current QLC drives are total shit in terms of steady-state performance. First 10-100 GB of data is written into the SLC cache which is fast, but then the drive switches to its QLC memory and even the linear write performance drops to ~90 MB/s which is actually worse than with HDDs! So, try to run a long linear write test and check the performance after writing a lot of data.
Last monday I performed a quick test with those two disks already, probably not that relevant, but posting it anyway:
But is random/sequential read performance still good? even during saturated write performance ? if so the tradeoff could fit quite some applications Sent from myMail for iOS Friday, 6 March 2020, 14.06 +0100 from vitalif <vitalif@yourcmc.ru>:
Hi,
Current QLC drives are total shit in terms of steady-state performance. First 10-100 GB of data is written into the SLC cache which is fast, but then the drive switches to its QLC memory and even the linear write performance drops to ~90 MB/s which is actually worse than with HDDs!
So, try to run a long linear write test and check the performance after writing a lot of data.
Last monday I performed a quick test with those two disks already, probably not that relevant, but posting it anyway:
ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
Hi, Sure, I will try the fio as requested by Marc Roos. I will be onsite again monday/tuesday. If you have a good way to test/show what you describe below, don't hesitate to post it here. I'll try it when I have the time. Are you saying that the write performance becomes bad (90MB/sec) for long lasting *continuous* writing? (after filling up a write buffer or such) But given time to empty that buffer again, it should again write with the normal higher speed? So in applications with enough variation between reading and writing, they could still perform good enough? MJ On 3/6/20 2:06 PM, vitalif@yourcmc.ru wrote:
Hi,
Current QLC drives are total shit in terms of steady-state performance. First 10-100 GB of data is written into the SLC cache which is fast, but then the drive switches to its QLC memory and even the linear write performance drops to ~90 MB/s which is actually worse than with HDDs!
So, try to run a long linear write test and check the performance after writing a lot of data.
Last monday I performed a quick test with those two disks already, probably not that relevant, but posting it anyway:
Yes. The drive treats some portion of cells as SLC, which having only two charge states is a lot faster and is used as cache. As with any cache-enabled drive, if that cache fills up either due to misaligned flush cycles or simply data coming in faster than it can flush, you’ll see a performance cliff. Also, uniquely for reasons I don’t yet understand, as the drive fills up, the size of that cache area decreases linearly until it hits a certain minimum. Thus an empty drive will have more cache in service than a 70% full drive. This article includes 8x10 color glossy pictures with circles and arrows and a paragraph on the back describing what each one is about. https://www.howtogeek.com/428869/ssds-are-getting-denser-and-slower-thanks-t... Which raises some interesting questions about the proper use of TRIM both by RBD clients and by ceph-osd, especially when dmcrypt is used. — aad
Are you saying that the write performance becomes bad (90MB/sec) for long lasting *continuous* writing? (after filling up a write buffer or such)
But given time to empty that buffer again, it should again write with the normal higher speed?
So in applications with enough variation between reading and writing, they could still perform good enough?
MJ
On 3/6/20 2:06 PM, vitalif@yourcmc.ru wrote:
Hi, Current QLC drives are total shit in terms of steady-state performance. First 10-100 GB of data is written into the SLC cache which is fast, but then the drive switches to its QLC memory and even the linear write performance drops to ~90 MB/s which is actually worse than with HDDs! So, try to run a long linear write test and check the performance after writing a lot of data.
Last monday I performed a quick test with those two disks already, probably not that relevant, but posting it anyway:
ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
participants (5)
-
Anthony D'Atri
-
Hermann Himmelbauer
-
jesper＠krogh.cc
-
mj
-
vitalif＠yourcmc.ru