Maximum IOPS for P4300?

[Tom Scott_2]

Hello, we have two P4300 Nodes with (8) SAS 15K drives each.. We are having performance issues in our VM environment and I want to exclude the SAN from any of the issues..

I've tried to find the maximum documented IOPS Total so I can compare that in my performance monitor but can't seem to find it anywhere?

Does anyone know? Also, does anyone know what the Queue Depth Total Average should be? I *think* it's the number of drives (8) x the number of nodes (2) x (2) if SAS so 8*2*2 = 32 (running network RAID10).

Thanks much!

[Steve Burkett]

Bryan over in this thread it should be about 800-960 IOPS per node, so about 1600-1920 IOPS for a 2 node cluster.

Jitun (who certainly seems to know his stuff) stated over in the bottom of this thread that the queue depth should be 1.5 to 2 times the number of SAS disks on the node multiplied by the number of nodes in the cluster to get the optimum performance out of your SAN, so yes, you got it right.

For comparison, on our reasonably loaded 3 node P4500 (12 x 15k SAS disk) cluster providing mostly vSphere datastores we're seeing averages at the moment of 1296 IOPS and 3.4 Queue Depth Total.

[Tom Scott_2]

Ok, so my 2 node cluster is averaging around 2,200 IOPS Total.. The average queue depth total is around 45. Both of these number are considerably over the number you provided..

As you are, we use our SAN for vSphere datastores.. In performance monitor in CMC I add all our volumes (Datastores) and add the IOPS total metric.. One of our datastore gets 2,100 IOPS total on average. On that datastore we have 9 *running* VM's..

I can't find in vSphere how to figure out which VM's are pushing what IOPS to try and possibly single out the culprit.. In this environment we are running vSphere 4.0.0. Anyone know if this is available anywhere?

Also, I appreciate the numbers given but is there an "official" number direct from HP in regards to IOPS and Queue depth?

Thanks much!

[Steve Burkett]

You could try something like the freebie Solarwinds Storage Response Time Monitor or get an eval of the vKernel vOperations Set to give you an idea on which VM is creating the most IOPS.

[bryan_hadzik]

G1/G2 and raid level have an impact on performance. Attached an image of the breakdown from an official HP planning guide.

[Tom Scott_2]

Where did you get that? Can you send me a copy of the file: tscott@fpfc.net

Also, I can't find anywhere the model number. I know it's a P4300 but unlike other HP's it isn't labeled with the G2 or G1.. 

I'm guessing it's a G1 since there is no "G" label? If it is a P4300 G1 I don't see a SAS model listed in your image below.

Thanks

[mc1903-1]

Hi Bryan,

Please can you advise which 'Official HP Planning Guide' you got these figures from..? I have been looking for IOPS figures for ages..!

I work for a HP Channel Partner and I have searched the public web site as well as Smart Portal, but I can't find them anywhere...

Thanks in advance...

Martin

[Jitun]

One will not find much difference in numbers between G1 and G2.

How many Number of VMs in a single data store is debatable.
It depends on how many VMs are IO intensive.
IO Intensive VMs should be spread across different datastores.

I would recommend 5-6 VMs per Volume.. But importantly have IO intensive Volumes spread to different datastores/volumes.