Block Storage - High Performance

[JacquesMrz]

As a user I want high performance Block Storage volumes available for my Public Cloud instances.

[desaintmartin]

Would you have an ETA for high perf block storage?

[Grounz]

what's the admissible iops for this new block storage volumes ?

[timscotland]

Seen on GRA9

[rgdev]

What is the official status on this ? Since it's available on GRA9 is it in beta ?

Edit: It also now shows up on the ovhcloud pricing

[rgdev]

I decided to give the high-speed-gen2 class of volume a spin on b2-15 on GRA9 and I consistently get around 6K IOPS, far from the max of 20K shown on the pricing above.

$ fio --name=rand-write --ioengine=libaio --iodepth=32 --rw=randwrite --invalidate=1 --bsrange=4k:4k,4k:4k --size=512m --runtime=120 --time_based --do_verify=1 --direct=1 --group_reporting --numjobs=4 --filename=/dev/sdc


rand-write: (g=0): rw=randwrite, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=32
...
fio-3.12
Starting 4 processes
Jobs: 4 (f=4): [w(4)][100.0%][w=23.4MiB/s][w=6000 IOPS][eta 00m:00s]
rand-write: (groupid=0, jobs=4): err= 0: pid=4568: Mon Sep 13 16:28:59 2021
  write: IOPS=6004, BW=23.5MiB/s (24.6MB/s)(2815MiB/120023msec); 0 zone resets
    slat (usec): min=4, max=4546, avg=21.18, stdev=13.60
    clat (usec): min=1097, max=42096, avg=21286.16, stdev=937.63
     lat (usec): min=1117, max=42108, avg=21308.87, stdev=937.66
    clat percentiles (usec):
     |  1.00th=[20055],  5.00th=[20317], 10.00th=[20579], 20.00th=[20841],
     | 30.00th=[21103], 40.00th=[21365], 50.00th=[21365], 60.00th=[21365],
     | 70.00th=[21365], 80.00th=[21627], 90.00th=[22152], 95.00th=[22152],
     | 99.00th=[22414], 99.50th=[22676], 99.90th=[27395], 99.95th=[32113],
     | 99.99th=[38011]
   bw (  KiB/s): min= 5856, max= 7200, per=25.00%, avg=6004.31, stdev=84.05, samples=960
   iops        : min= 1464, max= 1800, avg=1501.06, stdev=21.01, samples=960
  lat (msec)   : 2=0.01%, 4=0.06%, 10=0.06%, 20=0.52%, 50=99.35%
  cpu          : usr=2.25%, sys=4.38%, ctx=629120, majf=0, minf=35
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=100.0%, >=64=0.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.1%, 64=0.0%, >=64=0.0%
     issued rwts: total=0,720722,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=32

Run status group 0 (all jobs):
  WRITE: bw=23.5MiB/s (24.6MB/s), 23.5MiB/s-23.5MiB/s (24.6MB/s-24.6MB/s), io=2815MiB (2952MB), run=120023-120023msec

Disk stats (read/write):
  sdc: ios=65/719924, merge=0/0, ticks=56/15320424, in_queue=15293352, util=99.97%

$ fio --name=rand-write --ioengine=libaio --iodepth=32 --rw=randwrite --invalidate=1 --bsrange=4k:4k,4k:4k --size=512m --runtime=120 --time_based --do_verify=1 --direct=1 --group_reporting --numjobs=4 --filename=/dev/sdc


rand-write: (g=0): rw=randwrite, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=32
...
fio-3.12
Starting 4 processes
Jobs: 4 (f=4): [w(4)][100.0%][w=23.4MiB/s][w=6002 IOPS][eta 00m:00s]
rand-write: (groupid=0, jobs=4): err= 0: pid=4618: Mon Sep 13 16:47:21 2021
  write: IOPS=6004, BW=23.5MiB/s (24.6MB/s)(2815MiB/120023msec); 0 zone resets
    slat (usec): min=3, max=52519, avg=113.70, stdev=726.26
    clat (usec): min=747, max=73389, avg=21194.12, stdev=2392.99
     lat (usec): min=762, max=73412, avg=21309.34, stdev=2416.51
    clat percentiles (usec):
     |  1.00th=[14615],  5.00th=[16712], 10.00th=[20317], 20.00th=[20841],
     | 30.00th=[21103], 40.00th=[21103], 50.00th=[21365], 60.00th=[21365],
     | 70.00th=[21365], 80.00th=[21890], 90.00th=[22152], 95.00th=[22414],
     | 99.00th=[27657], 99.50th=[32113], 99.90th=[38011], 99.95th=[55313],
     | 99.99th=[66847]
   bw (  KiB/s): min= 5192, max= 7424, per=25.00%, avg=6003.94, stdev=191.74, samples=960
   iops        : min= 1298, max= 1856, avg=1500.96, stdev=47.93, samples=960
  lat (usec)   : 750=0.01%, 1000=0.01%
  lat (msec)   : 2=0.02%, 4=0.07%, 10=0.20%, 20=5.92%, 50=93.73%
  lat (msec)   : 100=0.06%
  cpu          : usr=2.08%, sys=3.75%, ctx=534121, majf=0, minf=36
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=100.0%, >=64=0.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.1%, 64=0.0%, >=64=0.0%
     issued rwts: total=0,720717,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=32

Run status group 0 (all jobs):
  WRITE: bw=23.5MiB/s (24.6MB/s), 23.5MiB/s-23.5MiB/s (24.6MB/s-24.6MB/s), io=2815MiB (2952MB), run=120023-120023msec

Disk stats (read/write):
  sdc: ios=0/719927, merge=0/0, ticks=0/14136052, in_queue=14121700, util=100.00%

$ fio --name=rand-write --ioengine=libaio --iodepth=32 --rw=randwrite --invalidate=1 --bsrange=4k:4k,4k:4k --size=512m --runtime=120 --time_based --do_verify=1 --direct=1 --group_reporting --numjobs=4 --filename=/dev/sdc


rand-write: (g=0): rw=randwrite, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=32
...
fio-3.12
Starting 4 processes
Jobs: 4 (f=4): [w(4)][100.0%][w=23.5MiB/s][w=6004 IOPS][eta 00m:00s]
rand-write: (groupid=0, jobs=4): err= 0: pid=4646: Mon Sep 13 16:54:39 2021
  write: IOPS=6004, BW=23.5MiB/s (24.6MB/s)(2815MiB/120022msec); 0 zone resets
    slat (usec): min=4, max=54046, avg=143.77, stdev=819.75
    clat (usec): min=1991, max=76944, avg=21163.85, stdev=2797.59
     lat (usec): min=2006, max=76965, avg=21309.17, stdev=2825.26
    clat percentiles (usec):
     |  1.00th=[11207],  5.00th=[16188], 10.00th=[20317], 20.00th=[20841],
     | 30.00th=[21103], 40.00th=[21103], 50.00th=[21365], 60.00th=[21365],
     | 70.00th=[21365], 80.00th=[21627], 90.00th=[22152], 95.00th=[22414],
     | 99.00th=[31327], 99.50th=[32375], 99.90th=[45351], 99.95th=[60031],
     | 99.99th=[69731]
   bw (  KiB/s): min= 4856, max= 7232, per=25.00%, avg=6004.23, stdev=255.53, samples=960
   iops        : min= 1214, max= 1808, avg=1501.04, stdev=63.88, samples=960
  lat (msec)   : 2=0.01%, 4=0.11%, 10=0.22%, 20=7.64%, 50=91.95%
  lat (msec)   : 100=0.08%
  cpu          : usr=2.01%, sys=4.10%, ctx=523546, majf=0, minf=38
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=100.0%, >=64=0.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.1%, 64=0.0%, >=64=0.0%
     issued rwts: total=0,720719,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=32

Run status group 0 (all jobs):
  WRITE: bw=23.5MiB/s (24.6MB/s), 23.5MiB/s-23.5MiB/s (24.6MB/s-24.6MB/s), io=2815MiB (2952MB), run=120022-120022msec

Disk stats (read/write):
  sdc: ios=0/719921, merge=0/0, ticks=0/13784141, in_queue=13763700, util=100.00%

[jbgomond]

What is the official status on this ? Since it's available on GRA9 is it in beta ?

Edit: It also now shows up on the ovhcloud pricing

Yes it seems to be on the princing page, is it stable ?

[JacquesMrz]

Hi all, here is the official status on this. New Block Storage Volume type is available in production in GRA9 region : high-speed-gen2 Allocation of the performance is done as follow: 30 IOPS allocated per GB within the limit of 20000 IOPS per Volume 0,5 MB/s allocated per GB within the limit of 1GB/s per Volume Price per GB is 0,08 € HT/month/GB

We understand many customers want to keep the old volume type available as they want average performance with small volumes. We currently don’t target to replace high-speed by high-speed-gen2.

We are working on the Website update to to make this crystal clear for everyone.

[desaintmartin]

That is great! Would you have an ETA for other zones?

[rgdev]

Thank you for the update. The test I published above was on 200GB volumes so it now makes sense why I only got up to 6K IOPS (30 * 200 = 6K). I'm also interested in knowing if those volumes are coming in other regions.

[roybj]

When it will be available on all zones ? I don't see it available in BHS ?

[telenieko]

@JacquesMrz:

(...) Allocation of the performance is done as follow: 30 IOPS allocated per GB within the limit of 20000 IOPS per Volume 0,5 MB/s allocated per GB within the limit of 1GB/s per Volume Price per GB is 0,08 € HT/month/GB (...) We are working on the Website update to to make this crystal clear for everyone.

I cannot find this information on the website, neither can I find the same for the other volume types. Where can I find that?

thx

[ZuSe]

@JacquesMrz any information when it will be available in GRA7 or SBG5? Kubernetes Clusters are unable to profit from the new storage atm as they cant be spawned in GRA9

[telenieko]

This issue was moved to "Released" by @JacquesMrz but it does not appear to be available on all locations. At least I can't find in on DE1:

+--------------------------------------+------------+-----------+
| ID                                   | Name       | Is Public |
+--------------------------------------+------------+-----------+
| 00fc334f-c1de-4705-bb88-0a58271b38c8 | classic    | True      |
| 8ab444e9-7206-4801-91bf-6e732e697de6 | high-speed | True      |
+--------------------------------------+------------+-----------+

What is the avaialability status of high-speed-gen2 per locations?

[arcalys]

On our end it is available on GRA9 and SBG7 only

[dave-b-code]

Now that managed kubernetes is available in GRA9, will we see this block storage option available soon in kubernetes? Thanks

[frabe1579]

In the website for Italian language, the details of the IO is wrong. It is: "330 IOPS/GB nel limite di 20.000 IOPS per volume" Should be: "30 IOPS/GB nel limite di 20.000 IOPS per volume" I think it's just a typo, should you take care of this? https://www.ovhcloud.com/it/public-cloud/block-storage/

[mhurtrel]

Hi @frabe1579 and thanks for the spot, indeed the right value is 30. I raised this to our marketing colleagues so that it is corrected quickly.

[frabe1579]

Thank you @mhurtrel .

[cambierr]

Hi @mhurtrel

Any plan for higher iops volumes ? 20k max + the iops/GB restriction are quite limiting DB & more general data-intensive usages

[JacquesMrz]

Ho @cambierr, copy that. We are currently working on a NVMe over fabric Block Storage solution, which should answer your needs. If everything goes well, BETA will start in the coming months.

[JacquesMrz]

S3 High Perf will be available at GRA next week !

[jbgomond]

Hi @JacquesMrz,

Do you have an ETA when it will be available in SBG5 ? Also, is it possible to keep this table updated ? It's very useful to follow what is available where ... but it seems it was not updated :(

Thanks

[qualifio-infrastructure]

Still not available outside of GRA, still no new information? Is this still even in the plans?

[laurentberger-ovhcloud]

You can follow availability here https://www.ovhcloud.com/en-ie/public-cloud/regions-availability/ When the service is deployed in a new region we update it. The next one to be opened is WAW1.

[revolunet]

Hi, looks like "High speed Gen2 block storage" is available in Gravelines, fine, but is it available inside managed kubernetes ? kubectl get storageclass only report csi-cinder-high-speed and its displayed as "Gen 1" in the OVH web UI.

any way to use gen2 in a kubernetes Peristent volume ?

[rgdev]

@revolunet If the region supports it you can add a storage class yourself :

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: csi-cinder-high-speed-gen2
parameters:
  availability: nova
  fsType: ext4
  type: high-speed-gen2
allowVolumeExpansion: true
provisioner: cinder.csi.openstack.org
reclaimPolicy: Delete
volumeBindingMode: Immediate

[revolunet]

Thanks @rgdev ! For some reason, when provisionning a 10Gi PVC, it looks very slow

ex : time dd if=/dev/zero of=/data/test2.img bs=1G count=1

1+0 records in
1+0 records out
real	2m 57.99s
user	0m 0.00s
sys	0m 3.10s

[rgdev]

IOPS scales up based on capacity, try instancing a 100Gi or 300Gi volume to see how it goes

[revolunet]

Thanks; some tests :

PVC 300Go high-speed-gen2 :
# time dd if=/dev/zero of=/data/test2.img bs=1G count=1
real	0m 7.10s
sys	0m 2.83s

PVC 1000Go high-speed-gen2 :
# time dd if=/dev/zero of=/data/test2.img bs=1G count=1
real	0m 4.94s
sys	0m 2.47s

(tested on b2-7 instances)

[LuckyFrost]

When will it be available in us?