« Welcome Scott Baker and Virtualutopia | Main | Wikibon Survey - VMware Storage Integration »

April 08, 2011


David Robertson

Great post Vaughn! one thing I will note on the Linux side of the house, if you are using LVM (Logical Volume Manager) or Oracle's ASM, as long as you use the entire device i.e /dev/sda and do not put a partition table down i.e /dev/sda1 then you will not be mis-aligned. Its only when there is a partition table layed down the be OS when you are misaligned.


Definetly a great "add-on" to your original post I am wondering what kind of impact could have the re-aligment of vmdks on large architecture.
As you correctly state "Be sure you understand the data flow before embarking on this last step.".

Parikshith Reddy

This is very good information...Can someone please advice if PLATESPIN actually does align the disks after a P2V? Thank You!

Nick Howell

alignment info on gparted for Linux: They've recently updated gparted to allow you to switch to a 1MB offset as well.


Aaron Delp

Hey Vaughn - First off, thanks for the link! A quick point of emphasis to your step 4. When you say you can migrate the application to a 2008 instance, you mean a fresh install (that is aligned by default) of 2008. You can't upgrade from 2003 to 2008 and gain alignment. If you were unaligned in 2003, you would be unaligned if you upgrade the vm to 2008 because it doesn't rewrite the partition tables. Makes sense when you think about it but I have seen that be a point of confusion in the past.



Will any of the new tools address thin provisioning after alignment? Correct me if I'm wrong, but if you have a thin provisioned VM that needs alignment, you not only have to take the VM offline to perform mbralign, but afterwards you would have to storage migrate it to a different datastore and then back? Without the storage migration step to re-thin provision it, I will make all my thin VMDKs "fat". Frankly, that's a PITA and is what has held us back from correcting all of ours. :)

I'm liking your idea of just migrating them to new 2008 servers. Lots of other benefits to go with it, like being able to resize the system partition with diskpart on the fly.


I'm pretty sure you can actually get the most recent version of Converter from VMware to produce aligned VMs -- you need to precreate the vmdk with an aligned partition and then point VMware Converter at an existing vmdk rather than creating one from scratch. A bit tedious but feasible (and not too bad if you create the vmdk thin and copy it as needed).

Peter Eicher

NetApp Syncsort Integrated Backup or NSB -- the joint data protection solution from Syncsort and NetApp -- also provides built-in P2V migration capabilities. Couple of cool things about it.

1. It starts with backups of your physical servers, stored as NetApp snapshots.

2. When the P2V migration takes place, it first uses the snapshot to boot the VM (creates a FlexClone). This means that your "conversion" time is about 5-10 minutes. That is, the new VM is up and running in that time.

3. The migration of the data to the VMDK takes place behind the scenes after the VM is already running off the FlexClone.

4. When the data is all moved to the VMDK, it invokes Storage vMotion to switch from the FlexClone to the VMDK. Zero additional downtime.

5. As part of the data migration from P2V, storage alignment takes place. So your new VMDK is correctly aligned.

6. When the server is migrated, data protection is already in place and backups just continue to run.

7. If needed, NSB can also migrate systems back to physical servers. V2V is also available.

All this is included as part of the licensing cost. There is no additional fees for the migration tools. And it's capacity based so you can migrate all the systems you want.

Note: primary storage does NOT need to be NetApp. Works with any DAS or SAN storage. Supports Windows and Linux systems.

It's pretty neat stuff!

Peter Eicher


ZFS misalignment due to variable block size is the pain point we're seeing now on our Netapp storage. The VMware side of things has been sorted for some time.

Vaughn Stewart

@All - thanks for the great dialog & feedback

@David - thanks for the additional info

@Alfwebcom - I'll see if I can get to this request, but it may be beyond my scope.

@Parikshith - yes platespin aligns vmdks

@Nick - great pint on gparted

@Aaron - right!

@Brian - MBRAlign will preserve the thin attribute with the exception of using I/O offload with Data Ontap 8.0.1. Now, I'd suggest thin or thick provisioning is a non-issue if you are using data deduplication... you are using dedupe aren't you? :)

@Andrew - good points, but may be a bit difficult for mass adoption. I have found if it isn't easy than most wont adopt.

@Peter - thanks for the info on Syncsort

@Tm - NetApp arrays don't use ZFS, so may I ask you to clarify your statement?


What about the problems in aligning Windows Server 2008 pointed out here:




@David: While not using a partition table works for ASM and maybe LVM, this is not supported by SnapManager for Oracle on RedHat.

And, how does LVM make sure that it's logical partitions (or filesystems) are aligned?


If you use Hyper-V I'd add one additional step: Use fixed VHDs. Dynamic VHDs insert container metadata inline with the filesystem data (as the VHD grows) resulting in misalignment regardless of your partition layout. So if you're fixing partition misalignment on a VM with dynamic VHDs make sure you also migrate (not convert) to fixed VHDs as well.


I thought I read that if you created the VM using vCenter, then vCenter will automatically align them for you. Is that correct?

Carl Skow

I noticed there is not much conversation around aligning VMs that exist in ESXi, just using the host utilities for ESX or third party tools that barely work on ESXi. I take it this is some of the NDA stuff? :(


@Malhoit: I'm afraid that's incorrect. It would be great if ESX had insight into how the guest OS writes data to disk and how the underlying storage handles disk blocks, but it doesn't, and that's why aligning is such a hassle. The only hope is that newer OS releases are aware of virtualization unlike the previous generation, so things are bound to get better once we start upgrading our old Windows 2003 and older installations.

Dale Wickizer

Great blog, Vaughn. Thank you!

David Robertson

As a update, a co-worker pointed out that ASM does require a partition table but LVM you can use a raw device. Thanks @wayne and @allen for pointing out my error

Jonathan Adair

Great post Vaughn. Really appreciate the insight!


Great insight Vaughn, thank you.

Vaughn Stewart

@Dave - thank for this, I'll forward it on to the engineers

@Carl - as soon as I can share what is going on with ESXi, I will.


@Vaughn - a few of us have noticed the mbrscan and (the newer) nfsstat -d output are not agreeing on the state of alignment - can you weigh in on if this just an interpretation of nfsstat -d output issue or if we actually have unaligned IO happening on what mbrscan says are aligned VMs?



Vaughn Stewart

@Fletcher - I don't have intimate knowledge around the data you are seeing with nfsstat, but I'd suggest you have an application in your VM that is creating a number of writes that are less than 4KB in size.

If what I suggest is accurate there's no need to be concerned, as it is normal behavior for the application.

I know you're well aware of misalignment, but please allow me to restate for those who may not be... What we want to avoid is having misaligned I/O for hundreds or thousands of VMs on an array. The inefficiency in I/O transfers due to a large mass of misaligned VMs will stress the array and lead to the eventual need to upgrade hardware.

Let me know if the small writes is or is not the case. I'd be happy to engage others to continue the conversation if needed.


The comments to this entry are closed.