[OmniOS-discuss] iSCSI poor write performance
Michael Rasmussen
mir at miras.org
Wed Sep 23 14:40:01 UTC 2015
Have you tried running iperf?
On September 23, 2015 4:06:53 PM CEST, "Martin Truhlář" <martin.truhlar at archcon.cz> wrote:
>Hi Hanno
>
>Thank you for your advice, unfortunatelly on dpool is ashift already
>set to 12 without any impact on performance.
>
>Martin
>
>dpool:
> version: 5000
> name: 'dpool'
> state: 0
> txg: 423442
> pool_guid: 8301756920046328435
> hostid: 390978448
> hostname: 'archnas'
> vdev_children: 6
> vdev_tree:
> type: 'root'
> id: 0
> guid: 8301756920046328435
> create_txg: 4
> children[0]:
> type: 'mirror'
> id: 0
> guid: 1196673952777635344
> metaslab_array: 34
> metaslab_shift: 33
> ashift: 12
> asize: 1000191557632
> is_log: 0
> create_txg: 4
> children[0]:
> type: 'disk'
> id: 0
> guid: 8124964091934866578
> path: '/dev/dsk/c1t50014EE00400FA16d0s0'
> devid: 'id1,sd at n50014ee00400fa16/a'
> phys_path: '/scsi_vhci/disk at g50014ee00400fa16:a'
> whole_disk: 1
> DTL: 490
> create_txg: 4
> children[1]:
> type: 'disk'
> id: 1
> guid: 9348868466535755709
> path: '/dev/dsk/c1t50014EE2B40F14DBd0s0'
> devid: 'id1,sd at n50014ee2b40f14db/a'
> phys_path: '/scsi_vhci/disk at g50014ee2b40f14db:a'
> whole_disk: 1
> DTL: 489
> create_txg: 4
> children[1]:
> type: 'mirror'
> id: 1
> guid: 9943497592636049032
> metaslab_array: 38
> metaslab_shift: 33
> ashift: 12
> asize: 1000191557632
> is_log: 0
> create_txg: 34
> children[0]:
> type: 'disk'
> id: 0
> guid: 2705367364579591435
> path: '/dev/dsk/c1t50014EE05950B131d0s0'
> devid: 'id1,sd at n50014ee05950b131/a'
> phys_path: '/scsi_vhci/disk at g50014ee05950b131:a'
> whole_disk: 1
> DTL: 488
> create_txg: 34
> children[1]:
> type: 'disk'
> id: 1
> guid: 5412107877453931054
> path: '/dev/dsk/c1t50014EE2B5E5A6B8d0s0'
> devid: 'id1,sd at n50014ee2b5e5a6b8/a'
> phys_path: '/scsi_vhci/disk at g50014ee2b5e5a6b8:a'
> whole_disk: 1
> DTL: 487
> create_txg: 34
> children[2]:
> type: 'mirror'
> id: 2
> guid: 4337686502023930092
> whole_disk: 0
> metaslab_array: 40
> metaslab_shift: 33
> ashift: 12
> asize: 1000191557632
> is_log: 0
> create_txg: 65
> children[0]:
> type: 'disk'
> id: 0
> guid: 12065653943105190290
> path: '/dev/dsk/c1t50014EE05958C51Bd0s0'
> devid: 'id1,sd at n50014ee05958c51b/a'
> phys_path: '/scsi_vhci/disk at g50014ee05958c51b:a'
> whole_disk: 1
> DTL: 486
> create_txg: 65
> children[1]:
> type: 'disk'
> id: 1
> guid: 7956964322079560255
> path: '/dev/dsk/c1t50014EE0595617ACd0s0'
> devid: 'id1,sd at n50014ee0595617ac/a'
> phys_path: '/scsi_vhci/disk at g50014ee0595617ac:a'
> whole_disk: 1
> DTL: 482
> create_txg: 65
> children[3]:
> type: 'mirror'
> id: 3
> guid: 13515811785015942389
> metaslab_array: 43
> metaslab_shift: 33
> ashift: 12
> asize: 1000191557632
> is_log: 0
> create_txg: 119
> children[0]:
> type: 'disk'
> id: 0
> guid: 2010958773514461606
> path: '/dev/dsk/c1t50014EE0AEAE7540d0s0'
> devid: 'id1,sd at n50014ee0aeae7540/a'
> phys_path: '/scsi_vhci/disk at g50014ee0aeae7540:a'
> whole_disk: 1
> DTL: 484
> create_txg: 119
> children[1]:
> type: 'disk'
> id: 1
> guid: 6920452460884353416
> path: '/dev/dsk/c1t50014EE0AEAE9B65d0s0'
> devid: 'id1,sd at n50014ee0aeae9b65/a'
> phys_path: '/scsi_vhci/disk at g50014ee0aeae9b65:a'
> whole_disk: 1
> DTL: 491
> create_txg: 119
> children[4]:
> type: 'mirror'
> id: 4
> guid: 13450996153705674574
> metaslab_array: 45
> metaslab_shift: 30
> ashift: 9
> asize: 120020795392
> is_log: 1
> create_txg: 172
> children[0]:
> type: 'disk'
> id: 0
> guid: 642840549260709901
> path: '/dev/dsk/c1t55CD2E404B88ABE1d0s0'
> devid: 'id1,sd at n55cd2e404b88abe1/a'
> phys_path: '/scsi_vhci/disk at g55cd2e404b88abe1:a'
> whole_disk: 1
> DTL: 494
> create_txg: 172
> children[1]:
> type: 'disk'
> id: 1
> guid: 17473204952243782915
> path: '/dev/dsk/c1t55CD2E404B88E4CFd0s0'
> devid: 'id1,sd at n55cd2e404b88e4cf/a'
> phys_path: '/scsi_vhci/disk at g55cd2e404b88e4cf:a'
> whole_disk: 1
> DTL: 493
> create_txg: 172
> children[5]:
> type: 'mirror'
> id: 5
> guid: 6461803899340698053
> metaslab_array: 520
> metaslab_shift: 33
> ashift: 12
> asize: 1000191557632
> is_log: 0
> create_txg: 422833
> children[0]:
> type: 'disk'
> id: 0
> guid: 15790186799979059305
> path: '/dev/dsk/c1t50014EE0AEABB8E7d0s0'
> devid: 'id1,sd at n50014ee0aeabb8e7/a'
> phys_path: '/scsi_vhci/disk at g50014ee0aeabb8e7:a'
> whole_disk: 1
> create_txg: 422833
> children[1]:
> type: 'disk'
> id: 1
> guid: 3033691275784652782
> path: '/dev/dsk/c1t50014EE0AEB44327d0s0'
> devid: 'id1,sd at n50014ee0aeb44327/a'
> phys_path: '/scsi_vhci/disk at g50014ee0aeb44327:a'
> whole_disk: 1
> create_txg: 422833
> features_for_read:
> com.delphix:hole_birth
> com.delphix:embedded_data
>
>
>-----Original Message-----
>From: Hanno Hirschberger [mailto:hannohirschberger at googlemail.com]
>Sent: Wednesday, September 23, 2015 2:43 PM
>To: omnios-discuss at lists.omniti.com
>Subject: Re: [OmniOS-discuss] iSCSI poor write performance
>
>Hi Martin,
>
>On 23.09.2015 10:51, Martin Truhlář wrote:
>> Tests revealed, that problem is somewhere in disk array itself.
>
>are you familiar with the ashift problem on 4k drives? My best guess
>would be that the 1 TB WD drives are emulating a block size of 512
>bytes while using 4k sectors internally. OmniOS is using a ashift value
>of 9 then to align the data efficiently (on 512 byte sectors!). This
>slows the whole pool down - I had the same problem before. The ashift
>value has to be 12 on 4k drives!
>
>Try the command 'zdb' to gather the values for your drives. Look for
>'ashift: 9' oder 'ashift: 12'.
>
>Regards,
>
>Hanno
>_______________________________________________
>OmniOS-discuss mailing list
>OmniOS-discuss at lists.omniti.com
>http://lists.omniti.com/mailman/listinfo/omnios-discuss
>_______________________________________________
>OmniOS-discuss mailing list
>OmniOS-discuss at lists.omniti.com
>http://lists.omniti.com/mailman/listinfo/omnios-discuss
--
Sent from my Android phone with K-9 Mail. Please excuse my brevity.
----
This mail was virus scanned and spam checked before delivery.
This mail is also DKIM signed. See header dkim-signature.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://omniosce.org/ml-archive/attachments/20150923/fe8e2556/attachment.html>
More information about the OmniOS-discuss
mailing list