Share some dedupe numbers?

[lohelle]

I have done a small dedupe-test in my environment. 4KB block size.

Server 1 DNS server, 2008R2, 8.8GB used of 40GB (thin vmdk)
Server 2 DNS server2, 2008R2, 6.8GB used of 40GB (thin vmdk)
Server 3 Web server, 2008R2, 7.4GB used of 40GB (thin vmdk)
Server 4 AD server, 2008R2, 8.4GB used of 40GB (thin vmdk)

Cloned server 1 to datastore. Spdata size = 7 GB
Cloned server 2 to datastore, Spdata size = 7.2 GB
Cloned server 3 to datastore, Spdata size = 8.2 GB
Cloned server 4 to datastore, Spdata size = 10.1 GB

All servers installed from ISO. Not cloned or not installed from templates.

31,4GB used on datastore. 10.1 GB spdata, 900MB spmetadata, 315MB spbitmap

11.3GB vs 31.4GB is quite good. Could be some nice savings if I put all my
low-IO 2003 servers on one LUN and all 2008 R2's on a different LUN.

I have not tested higher IO servers yet. But my 8-core 2,3 GHz Opteron server easily saturate the gigabit link used during cloning.

Would be nice with HA + dedupe!

[lohelle]

Feature requests:
Vaai support (reclaim of space when deleting files in VMFS)
Dedupe support for HA-devices
And a general request:
Change sync-channel settings on HA-target

[lohelle]

Strange...
I created a NTFS-partition from a 800GB dedupe device (4KB block). I installed a Windows Server 2008 R2 Hyper-V VM. Then I installed second VM the exact same way.
Dedupe size almost doubled.

Then cloning in Vsphere the size was almost unchanged (different servers, not two of the same VM's)

[anton (staff)]

VAAI should be part of V6 (and we'll have own intellectual path to return back unused space).

Dedupe support for HA is also part of V6.

Not sure what you mean with "change sync-channel..."

Could you please spawn a dedicated thread with requests in Beta or here (I'll create a shadow copy in any case).

Thanks!

Feature requests:
Vaai support (reclaim of space when deleting files in VMFS)
Dedupe support for HA-devices
And a general request:
Change sync-channel settings on HA-target

[anton (staff)]

Offsets floated in some way. What cluster size for NTFS do you use?

Strange...
I created a NTFS-partition from a 800GB dedupe device (4KB block). I installed a Windows Server 2008 R2 Hyper-V VM. Then I installed second VM the exact same way.
Dedupe size almost doubled.

Then cloning in Vsphere the size was almost unchanged (different servers, not two of the same VM's)

[lohelle]

4KB, same as dedupe block size.
Created as a GPT disk.

Not important as I will use vSphere only, but just a bit strange.

I'm going to do some more dedupe tests in a few days with 32 SAS-disks in RAID 10 + infiniband to ESXi host.

[anton (staff)]

Well, quite important for us as we'd like to pinpoint and fix the issue

4KB, same as dedupe block size.
Created as a GPT disk.

Not important as I will use vSphere only, but just a bit strange.

I'm going to do some more dedupe tests in a few days with 32 SAS-disks in RAID 10 + infiniband to ESXi host.

[lohelle]

Any test you would like me to do? Sector sizes and block sizes? (I was referring to the hyper-v issue)

[anton (staff)]

Could you please clarify (maybe in e-mail) what exactly you did so we could find out why dedupe did not actually work for second image. Thank you!

Any test you would like me to do? Sector sizes and block sizes? (I was referring to the hyper-v issue)

[lohelle]

I have a new server that will replace node 2 in our HA setup.

I installed Starwind 5.8 and created a 800GB dedupe-disk with default options. I then used MS Iscsi initiator on the same server to connect to the new target.
Converted the disk to GPT and formatted as NTFS with default options. Checked, and cluster size is 4K, same as block size for the dedupe-disk.

I added hyper-v role and created a new VM (80gig disk). Installed from Windows Server 2008 R2 iso. Default install. Did nothing after install was finished.
I added a second VM with EVERYTHING the same except for the name (added _2)

Disk usage inside both VM's = 10GB
Dedupe disk file after first VM install = 6,7GB
Dedupe disk file after second VM install = 12 GB

[Vitalii (staff)]

Could you try using 512 block of deduplication for that purpose?

It could be the problem of Hyper-V file format that does not necessarily align data to 4k boundary, so the dedupe engine could not match these blocks.

[lohelle]

512b dedupe (200GB disk)
5.5GB usage (just spdata) after 1 VM install. 6.7GB after the second. But about 10GB of metadata and bitmap also.. and bad performance.. and high mem usage.

Tried 4GB block again and GPT disk. Formatted and actually selected 4kb clusters (even if this was used automaticly last time)
Same results as previous 4kb results.

Same results with MBR and 4kb block/4kb cluster.
As far as i know, Windows Server 2008 R2 setup should align partitions automaticly? (I was thinking about inside the VM's)

[anton (staff)]

Yes, it's VHD align issue. Confirmed! We'll check what could be done to fix / workaround this.

512 byte dedupe is not intended for production actually (only experiments). Load is too heavy.

512b dedupe (200GB disk)
5.5GB usage (just spdata) after 1 VM install. 6.7GB after the second. But about 10GB of metadata and bitmap also.. and bad performance.. and high mem usage.

Tried 4GB block again and GPT disk. Formatted and actually selected 4kb clusters (even if this was used automaticly last time)
Same results as previous 4kb results.

Same results with MBR and 4kb block/4kb cluster.
As far as i know, Windows Server 2008 R2 setup should align partitions automaticly? (I was thinking about inside the VM's)