[sheepdog-users] The iscsi performance

Hitoshi Mitake mitake.hitoshi at lab.ntt.co.jp
Mon Mar 30 10:40:13 CEST 2015


At Mon, 30 Mar 2015 11:05:40 +0800 (CST),
李文涛 wrote:
> 
> [1  <multipart/alternative (7bit)>]
> [1.1  <text/plain; GBK (base64)>]
> When I test the performance without option  '-n'.
> From the client to server's disks, I analyze the client cpu, the iscsi networks environment, the server cpu, the server disks.
> The client have more than 50% wait, the rest is idle.
> The iscsi networks environment is via 10GbE.
> The server cpu is more than 95% idle.
> The server disks is just 5% utilization.
> So I means can sheepdog make maxmize use of the hardware?
> And my environments have much more hardware resource for sheepdog to use.

Could you provide the detail of your environment? e.g. how many sheep nodes, iSCSI initiator and its OS, fio parameters, network topology, etc.

Thanks,
Hitoshi

> 
> 
> 
> 
> 
> At 2015-03-30 10:12:41, "李文涛" <liwentao988 at 163.com> wrote:
> Hi Valerio,
> Sorry for reply so late.
> I saw the option '-n' is 'drop O_SYNC for write of backend'. I think it can only improve write performance.And it could make my data danger.
> And have you test the performance? How are them?
> My cluster is 2 nodes.And the copies is 2.
> 
> [root at server240 ~]# dog cluster info
> Cluster status: running, auto-recovery enabled
> Cluster created at Fri Mar 20 18:13:00 2015
> Epoch Time Version
> 2015-03-27 15:15:33 131 [192.168.200.227:7000, 192.168.200.240:7000]
> 
> [root at server240 ~]# dog vdi list
> Name Id Size Used Shared Creation time VDI id Copies Tag
> dog8 0 100 GB 100 GB 0.0 MB 2015-03-27 10:26 e94b23 2
> 
> At 2015-03-28 18:12:21, "Valerio Pachera" <sirio81 at gmail.com> wrote:
> 
> 
> 
> 
> 
> 2015-03-27 11:17 GMT+01:00 李文涛 <liwentao988 at 163.com>:
> 
> Have sameone test the performance of iscsi ?
> 
> 
> 
> Is sheep daemon running with option '-n' ?
> 
> If not, stop your cluster and this option.
> 
> How many copies is you cluster using?
> 
> If you run 'dog cluster format' without argument it's keeping 3 copies.
> 
> If the cluster is small, you may use 2 copies and it will be faster.
> 
> 
> 
> [1.2  <text/html; GBK (base64)>]
> 
> [2  <text/plain; us-ascii (7bit)>]
> -- 
> sheepdog-users mailing lists
> sheepdog-users at lists.wpkg.org
> https://lists.wpkg.org/mailman/listinfo/sheepdog-users



More information about the sheepdog-users mailing list