latency on P4500 28.8

[bublik]

I would to expect approximately 2000 IOPS out of my system and maybe 160-180 MBps, am I wrong?

Hi,

I have P4500 P2 28.8 4 node storage cluster. networking is RAID 5 and Networking is RAID 10

flow control disabled. no jumbo frames on the iSCSI network implemented yet. Each node has 2 network cards bond

I am using IOMETER to do a performance test and receiving some horrified results or maybe I am not understanding something here. 

g disk is SQL log disk NTFS formatted with 64K

t disk is a test disk with default NTFS format.

f disk for SQL data NTFS formatted 64K

========================================

I have used SQLIO for testing an got different results.

There are recommendations from expert is to format NTFS disk with block size 64k for data and logs which I did. I am seeing a little better results for my T disk that is default size than G disk with 64K.

I would to expect approximately 2000 IOPS out of my system and maybe 160-180 MBps, am I wrong?

----------------------------------------------------------

I have used SQLIO for testing an got different results.

There are recommendations from expert is to format NTFS disk with block size 64k for data and logs which I did. I am seeing a little better results for my T disk that is default size than G disk with 64K.

[Davy Priem]

I would suggest you first configure (flow control, ...) your networking and than test again.

[bublik]

is Deleting Bond requires downtime? if yes, how long should I anticipate?

I have four nodes I have to delete the bond and reconfigure networking. But storage already running vmware in production.

[Bart_Heungens]

yes there will be short downtime per storage node, not on the volumes however...

if U do it 1 box at a time, and U wait untill the first box is back online and wait untill the volumes are re-sync'ed, it won't give U downtime on the volumes...

[bublik]

Someone has mentioned I might have a problem on the volume level when I do it on my last node... Any comments on this?

What If I remove the node from the cluster? It will start re-striping the data, right?

[Bart_Heungens]

If you have quorum, or a majority of managers running (with 4 nodes you need at least 3 managers running) all volumes stay online...

So if you do it node per node, and you wait that they are back online and sync'ed, the volume stays online... This is the same with firmware/patch upgrades for instance...

Yes the re-sync'ing happens automatically...

No you don't need to remove the node from the cluster... If you do that, the full volume must be replicated when you add the node back to the cluster, which will take much more time...

[bublik]

Thank you. I will give that a shot.

[bublik · Accepted solution]

Before configuring jumbo frames and flow control I have tried to increase Outstanding OI in IOMETER to 96 which is double of the disks in the array. In this case I got great results.

[Jitun]

Outstanding IOs is Queue Depth Which should be 1.5 to 2 times the number of Disks (SAS Disks) on the Node multiplied by the Number of Nodes in the Cluster.Then you are providing optimum work to your SAN.