StarWind Software

Hi:

Installing SW ISCSI San Free Edition for the time being (considering paid version for near future) (thank you very much for the free version), installation to be performed on Windows 2012 Server with two different HyperV 2012 Servers on separate physical hardware accessing the SW server over Iscsi. Two 1 GB nics on each server.
We did a lot of research on this forum and other places about best practices but still we have some questions, any comments will be appreciated.

This is what we understood is recommended.
1. If using deduplication cluster size of 4k gets better results.
2. Deduplication recommended for backups but not for VM's (at least not until v8)
3. Don't use load balancing (I guess LACP is included) for bonding but MPIO instead.
4. Use NTFS
5. Disable write cache on the controller and leave only SW Write back cache

The questions we have and which is unclear for us is:
1. Should read ahead cache on the controller be disabled as well (we are using Dells were you can enable it or have it adaptive read ahead were it activates automatically).
2. Should the controller RAID (container) be created with a block size matching the future block size to use on logical partitions and deduplicated block size? or doesn't really matters?
3. Which RAID would you recommend using? RAID5 + HS or RAID6, which one performs better for VM traffic?
4. For VM's the device to create and attach to a target with the ability to grow it in the future would be the virtual disk/image file device as the other devices won't allow extend the storage?
5. The Snapshot and CDP device looks nice but seems cannot be expanded nor it seems can the snapshots and cycles between snapshots can be changed later on.
6. This is really an important question, is it recommended from the HyperV initiators to map the exported iscsi targets as drives and then create the vhdx files in these OR link the iscsi exported targets directly into the HyperV VM's as hard drives bypassing the Windows Server traditional mapping? like vSphere does.
7. Should we use the SW Free Native HyperV product instead of the SAN free edition?

Any help and comments are appreciated

Thank you in advance

Luis

1) You may use RAID built-in cache if you already have one configured.
2) RAID stripe size (256KB+) will be always bigger compared to dedupe block size (4KB). But it's OK to have them of the different size.
3) Before V8 with LSFS you need to stick with OBR10 (One Big RAID10). With LSFS parity RAIDs are OK (for performance, rebuild time is still an issue so make sure you have a decent backups).
4) Grow is possible of course.
5) Don't use CDP in production now as this feature is going to be suppressed by LSFS.
6) Use hypervisor initiator to connect to iSCSI for performance, configure CSV, put VMs VHDX on them and go on. Do not use pass-thru disks as they create a lot of limitations.
7) Native SAN (vSAN in the future) is for the people willing to run storage virtualization stack on hypervisor. It's the upcoming trend (faster, cheaper and more reliable). You can use both ways.

Thank you

1. So it's ok to use controller's read cache but not write cache correct?
6. configure CSV (what's CSV)
7. Was under the impression centralized storage was actually better and cheaper to maintain, of course leaving the hypervisor to take care of redundancy using live migration and / or replication you can run local VM's and still achieve a lot of redundancy

When is v8 expected to be GA?

1) You can use both. But using write cache w/o BBU (or even with it) in a non-clustered environment is dangerous.

6) Clustered Shared Volume. See:

http://technet.microsoft.com/en-us/libr ... 59255.aspx

7) No, not any more.

Beta-1 is scheduled for this week with Beta-2 following in July and RC and release in August.

luisma wrote:Thank you

1. So it's ok to use controller's read cache but not write cache correct?
6. configure CSV (what's CSV)
7. Was under the impression centralized storage was actually better and cheaper to maintain, of course leaving the hypervisor to take care of redundancy using live migration and / or replication you can run local VM's and still achieve a lot of redundancy

When is v8 expected to be GA?