NetApp NFS datastore layouts

[shane_presley]

We're planning a NetApp NFS deployment for our 8 host VMware environment (100 VMs). The VMs are all infrastructure servers (low I/O), except one Exchange server.

Are there any best practices about how to layout the datastores?

Would one big datastore work for all 8 hosts? Or for performance reasons should we break them up into smaller chunks, possibly one per server.

Thanks

[RParker]

All depends on the disk (type).and how you setup the aggregates on Netapp.

We do one really big (28 / 56 disk) target. we have LUN's so we have to use smaller volumes of less than 2 TB. This would be fine for FC drives, not SATA. SATA has to be relatively small, because the IO is different.

But NFS in my opinion should be one big data volume, that should be fine.

[mike_laspina]

Hi,

NetApp recommends that you create separate datastores for temp and swap activity. This is very important for performance and disk consumption. If you leave pagefiles and temp file activity on snapshot based volumes you will consume large amouts of storage and fragment the WAFL file system.

Have a look at

Regards,

Mike

vExpert 2009

[shane_presley]

Thanks guys!

RParker --We will be using 24 450GB FC disks. So large NFS volumes should be ok? What specifically should we take into account when building the aggregates? We'll have a NetApp engineer helping us with us, I was just hoping to have a little background knowledge before he gets on site.

Mike -- Great blog, thanks. I hadn't thought of the temp/swap implications on snapshot volumes. Since our environment (100+ VMs) wasn't originally designed that way, I'll have to look into the implications.

[mike_laspina]

Hi,

You can simply create an additional vmdk on each VM and point the swap/temp files there. It can be transistioned before hand at your next patch window. Then you can migrate them using the advance option and separate the noise vmdk to the non-snapshot file system.

Regards,

Mike

vExpert 2009

[HendersonD]

Mike,

We have a single NFS datastore on our Netapp FAS3020 that houses about 35 VMs under VSphere. We do have a separate volume for vmware swap files and do not snapshot that volume.

I know that Netapp also recommends creating a separate volume for each Windows VM and placing the windows pagefile there and then excluding this volume from snapshots. When I delved into this a bit, it seemed like a lot of work to do this for 35 VMs. Is this a lot of work? Is it worth the effort?

Dave

[mike_laspina]

Hi Dave,

VMware swap files have very little activity when memory resources are available at the ESX host. In contrast Microsoft swap files are always active and will generate constant chatter. It is very benificial to move that I/O noise out of the snapshot protected storage area.

There are 4 basic reasons for this design.

1. The snapshot will hold junk I/O until the point in time is destroyed, which means you will need to a shorter time window rolling stragtegy to keep the consumption under control.

2. The constant chatter is 100% unique and bypasses any dedup benifit.

3. This type of I/O will accelerate the storage framentation rate which will increase maintenance cycle time.

4. Without separation you cannot isolate the I/O to a volume which does not tax the availble nvram. e.g iSCSI non O_sync dram cached volume.

Regards,

Mike

vExpert 2009