Subscribe

IOPS influences

Hi folks,

Can anyone here tell me what influence if any, Anti-Virus checking as described in: http://media.netapp.com/documents/tr-3107.pdf or clustering

have on Filer IOPS values?

Also is there any rule-of-thumb which can help determine what added IOPS advantage would be gained by adding a Pan-card to a certain system configuration?

Thanks in advance for your help!

Regards,

Mark

Re: IOPS influences

Hi

1).  AV will increase latency on CIFS as you have to request the file, which is scaned by a 3rd system and then past to the host.  There will also be douple the I/Os but in theory the 2nd read will come from cache

2).  Not sure 'clustering' you you mean active / active filer pair?  Or metro cluster or host cluster like MSCS

3).  Do you mean PAM - Performance Acceleration Module?  Yes they can help with READ I/Ops if your system is a supported type with the correct DoT version.  1st major improved system would be VDI boot up times.

Hope this helps

Bren

Re: IOPS influences

Hi Brendon,

Thanks for your response.

In this case it concerns a FAS3140 stretched act-act MetroCluster based solely on SATA disks (customer demand).

Due to the demand to run SATA disks we got into a major discussion on the number of required spindles etc in relation to the customer's IOPS requirement.

As usual they need more IOPS than they can pay for.

Anyhow as it stands with the number of shelfs we've configured we can give only ever so slightly more than what they require. So we need to know how much IOPS any additional processes such as virusscanning, cluster operations and anything else you can think of, might consume.

Yes I meant the accelleration modules. Question now being how many additional  IOPS's should we indicate or can we calculate with in our proposal to the customer?

regards,

Mark

Re: IOPS influences

The good thing though is that read performance on MetroCluster is slightly improved (vs. "normal" setup) as reads can go from both plexes.

I am not quite clear by how much this actually improve the read performance - in theory twice as many spindles are involved, but that's too good to be true

Regards,
Radek

Re: IOPS influences

It does not look like PAM is currently supported on metro so that can be ruled out.  I do not know when version 2 is due.  http://now.netapp.com/NOW/knowledge/docs/hardware/filer/perfaccel.pdf

I am a NetApp end user so I do not have access to technical information on the IO/s figures that you need.

Bren

Re: IOPS influences

Actually sys config guide says PAM cards [X1936A-R5] are OK to be put in a MetroClustered FAS3140:

http://now.netapp.com/NOW/knowledge/docs/hardware/NetApp/syscfg/scdot7311/index.htm

Re sizing tools:

CIFS Sizer doesn't show PAM card option (at least I can't see it), but Custom App Sizer does. None of them though mentions any possible impact of MetroCluster read performance improvement.

Regards,
Radek

Re: IOPS influences

Just to be all picky, improved read performance is actually a function of SyncMirror (something that you need as part of MetroCluster but is separate and can be used without MetroCluster).

Also, the option to service reads from both plexes of an aggregate is optional (it's turned off by default but can be enabled).