Defragging storage drive on an EVA8100

[Steve616]

I wonder whether anyone has any clear guidance on this.
Essentially, before our EVA we had an old MSA box, that we'd defrag the disks on a weekly basis (through whichever node was active), this constitutes roughly 4Tb of data.
This was split down into drives that performed staggered defrags after the backups early in the morning throughout the week. They took approx 40 mins to complete.

We've now upgraded to 2 x EVA8100s that replicate between 2 sites. (this is now consolidating more storage services into one, hence the big hike to the 8100.

I asked the installer about whether we should keep this practice going from the live front end and was told no as the EVA stores data in a scatter gun effect accross the entire disk set and so it's all over the place anyway.
That said, I know from experience that if you let it get out of hand from the OS point of view there will still be an increasing delay in access from a server response point of view as the O/S will still be piecing different fragments of files together adding unnecessary I/O overhead to the file access process.
I've also seen it where the disk goes into a "too fragmented" state where you then have to take the disk offline to defrag (definately unwanted on this amount of data).
Once I made this argument the response became a little hazy and I've still not really had a satisfactory response.

I've set it up on the live nodes again, and now it only takes 20mins each week to complete (this EVA is rapid to say the least) however, I'd still like some authoritive response on this?
I realise that this will have an affect on site to site replication as it is a byte-wise replication, but the frequency of defrag really makes the job far smaller in terms of bytes to alter and we are running Async replication.
If someone could offer a general "stance" as to fragmentation or their own experiences, i'd be most appreciative.

[TxMale]

Steve,

Since no one else has given you a response on your question I will give you my $0.02 on the subject.

What you were told is correct.

- The EVA will not benefit from a host OS defrag, nor will it be hurt by you doing one other than the increased I/O during the defrag.

- As you correctly pointed out, the Windows OS will still believe that it's files are being fragmented as time goes on.

- CA is a block level replication of the blocks that are changed "on the EVA vDisk"

It is somewhat unclear to me without a machine to play with whether doing a Windows OS level defrag will affect the amount of data (blocks) to be replicated between the sites.

As you observed the Windows OS defrag is happening in 20 minutes instead of apparently a much greater time for defragging the MSAs. One possibility that occurs to me is that the reading and writing that Windows does to defrag the files are all happening in cache and possibly not affecting the CA dirty block bit maps at all. You could test this by observing the replication link traffic before, during, and shortly after the defrag occurs.

I hope this helps...
Billy Beckworth
HP Storage Consultant
Thursday, November 19, 2009 10:42:35 AM