Large RDM LUN backing up white space

[dcs115]

I have a fileserver (Windows Server 2008 R2) that has a 1 TB RDM iSCSI LUN attached to it and am having a problem where B&R v6 is backing up the entire drive (including the empty white space), not just the used space.

What i am used to seeing on my other VM's is that only the used bits are backed up (ie i have a 40GB hard drive where my OS lives and B&R only backups up 20 GB of used space). Is this due to the fact that the Hard Disk is RDM, is there some sort of setting that i missed on this? I should note that B&R fails over to "Network Mode" when backing up, presumably because of the RDM drive.

[Gostev]

What exactly do you mean by "backing up the entire drive"? Do you mean it is reading the entire drive? Or the backup size is too large?

[dcs115]

That its reading the entire drive. I saw your first post and it sounds like RDM is the cause.

[Gostev]

Yes, I just wanted to make sure we are on the same page and I fully understand the issue before I go and blame RDM for it.

Essentially, one of many benefits of VMDK over RDM is that VMFS has the knowledge of which VMDK blocks are zeroed, so that the backup application does not need to read those blocks. There is special CBT query that we use during the full backup that returns only non-zeroed disk blocks, this allows our processing engine to only read actually used virtual disk blocks. Unfortunately, RDM disks do not provide this capability - we have to read the whole disk during the full backup. Yet another reason to switch to VMDK

[dcs115]

I used storage migration to convert my RDM's to VMDK's, but it still doesn't look like its working. The backup is still taking the entire 1TB of space even though the disk is is about 45% full. I used thin provisioning when i brought the disk over from RDM. Oddly enough, the "size" of the vmdk in vsphere is 750GB. B&R doesnt care about thin or thick provisioning, right? Via the CBT query you guys take, it should still just read the non-zeroed block? The vmdk lives on a VMSF5 disk, so block size isnt an issue. Is there anything else i can do to "force detect" the non-zeroed blocks?

I am going to try this (http://www.virtualizationteam.com/virtu ... space.html) to fix my thin provisioning problem and i will see if it fixes my CBT problem too.

[Gostev]

Via the CBT query you guys take, it should still just read the non-zeroed block?

This particular CBT query is very unreliable (does not work most of the times). There are existing topics discussing the use of CBT during full backup, and this particular issue in more details.

[dcs115]

Just so it is noted for the future. After converting an RDM to VMDK (thin provisioned), B&R could not detect used\unused blocks on the disk so it would backup the whole thing (like a normal RDM disk). In order to make it detect these blocks i had to follow the steps noted here (http://www.virtualizationteam.com/virtu ... space.html):

1. on your windows server, run sdelete -z driveletter:\ (-z is the option for zeroing blocks in the new version of sdelete, the old version was -c)
2. migrate the vdisk to a different datastore with a different block size

I did this on a 40GB test disk, but am going to try it on a production 1TB disk this weekend.