VM Alignment with NetApp and VMFS datastore

[thmadd]

In short, we have vsphere VMFS datastores connected via FC from a NetApp filer. The lun is 'lun type' VMWARE, and the VMFS was created from the vSphere client. According to the NetApp best practices for Alignment in a virtual alignment, by following these steps the guest vm's should be aligned. When I look at my partition offsets in the Windows servers they show the offset is set to 32,256 bytes. Is this VM's OS aligned, or does it need to be aligned?

[a_p_]

When I look at my partition offsets in the Windows servers they show the offset is set to 32,256 bytes. Is this VM's OS aligned, or does it need to be aligned?

This value refers to the guest OS. Starting with Windows Vista/Server 2008 MS changed this to 1 MB.

Take a look into the manual you mentioned for the mbralign utility.

André

[thmadd]

Correct about the 2008 servers and vista. I hae used the mbrscan and mbralign utility for all of our vm's that are sitting on our NFS datastores. What I'm unsure of is the vm's that are residing on our VMFS datastores, and whether or not they need to be aligned.

[a_p_]

Alignment is the same for physical and virtual systems. VMFS is only one more layer between the VM and the physical disk.

There are some samples/pictures in the NetApp alignment document which explain this.

If you really need to align each VM depends on the disk IO. The higher the disk IO, the more you could probably benefit from alignment.

André

[FranckRookie]

Hi Thmadd,

By doing this, you align the VMFS to the NetApp file system. But you do not align the file system inside the VM.

For that, you need to align it with Diskpart while installing the guest OS or use MBRAlign after Windows installation.

Regards

Franck

[thmadd]

Thanks for the input. I gather the difference between the VMFS and GOS alignment. However by using VMFS with unaligned GOS, is there any noticeable performance degradation? From what I can find in VMware's best practice for performance guides, there is no performance loss by having unaligned GOS on VMFS.

[FranckRookie]

Depends on what exactly does your VM.

If it handles only big files, you should not notice any difference. But if it works with a lot of very small files, you can gain up to 10% (from what I read...).

[J1mbo]

There are two very seperate issues here

- aligning to the underlying RAID stripe size, which provides no real performance penalty in terms of latency provided the storage is not fully loaded, as split requests (at the hardware) can be served in parallel (by the two physical disks involved). For example if the RAID set has 64K stripe, and the application is using 8K transfers, and the guest partition is misaligned at 31.5K as in this case, on average every 8th IO will straddle two spindles and hence add about 12.5% to storage IOPS load.

- aligning to the minimum addressable storage layer unit. Since this is NetApp, presumably this is NFS storage? Hence ESX(i) must work in the smallest unit which is probably 4K (not 512-byte sectors as for iSCSI). In this case the 31.5K offset results in every write being executed using read-update-write by ESX(i) (what else can it do?) giving a large performance impact.

Benchmarking your storage within your guests is the key, I wrote a little about using IOMeter for this here.

HTH

http://blog.peacon.co.uk

Please award points to any useful answer.