write latency

[tdubb123]

Can you tell if there is latency here? I am getting these spikes but only on 1 hba. I am use round robin, how come I dont see both hbas spiking at same time?

[Alistar]

Hello,

can you provide us with more info on this issue? How many HBAs do you have connected to the host? Are there any IO/throughput-intensive VMs working on that host? Do you observe the same latency on the array itself? Did  you check if the disks are busy on the array? This would help us get closer to the root cause of the issue.

The latency during the night could be because of backups running on the array and that is normal.

[jrmunday]

Hi tdubb123,

If you are using RR, then I would say it is odd to only see this on one HBA. I have seen similar behavior (EMX Symmetrix VMAX 40K) where a faulty SAN port caused issues with one specific path. I identified the path by running an iometer workload using a Fixed PSP. With this running I then systematically changed the preferred path to see how this impacted performance. One of my 4 paths performed poorly, while the others were fine. Once the FC was moved onto a new port everything was back to normal (RR performed as expected again).

Perhaps you have something similar?

Cheers,

Jon

[tdubb123]

hi

if its an active/active array, and we are using RR, should i change it to fixed instead? Seems like fixed is the preferred psp for active/active arrays?

[tdubb123]

hi I see these spikes consistent among all hosts in the cluster and I am not using blades. I dont think its specific to the san port on the hosts

[jrmunday]

The SAN hardware vendors should provide guidance on the recommended PSP. They normally provide a "best practice" document with guidance on configuration.

What array are you using?

[tdubb123]

I think its an active/active HP XP array

[jrmunday]

Looking at this HP XP7 technical white paper, the recommended best practice PSP is RR;

http://h20195.www2.hp.com/v2/getpdf.aspx/4aa5-2046enw.pdf