Questions Starwind Free Edition

[ticktok]

While setting up a Cluster Shared Volume and transferring an existing VHD of 20Gb to the cluster volume on a Windows 2008 R2 Datacenter Hyper-V in our development lab the host's grew by some 4Gb while caching the writes to the Starwind clustered target with a 3Gb cache and write-back caching strategy. The dialog box showing the file copy in Windows Explorer then remained at 5 seconds left as the 4Gb of cached writes was emptied from the Hyper-V host memory to the Starwind target.

My question is if I shut-down the Hyper-V host would it wait till the writes had been flushed before shutting down?
If the machine on which the Starwind target was shut down while a Hyper-V host has not flushed it's write cache, what happens to the unflushed write's still sitting in the Hyper-V's memory. Or is there some sort of communication between the Starwind target and the iSCSI clients that prevent the Starwind host machine shutting down and it waits till all iSCSI clients have flushed their caches?

Should the server on which the Starwind software has been installed be dedicated or can other Windows file services DFS, DFS Replication run?

With regard to a dedicated server just running Starwind target software for iSCSI clients on a 1Gb network (my clients have usually tight budgets and 10Gb network gear is too expensive) how could you achieve the best network throughput:-
1). MPIO and add extra NICs to both the Starwind server and its Hyper-V and other iSCSI clients
2). Link aggregation with a managed switch and add extra NICs to both the Starwind server and its Hyper-V and other iSCSI clients
3). Combination of the above

If it were not a dedicated server and other File services were running DFS, DFS Replication, SMB shares how would this affect the question above regarding maximising network throughput?

With regard to HYPER-V clustered volumes which is primarily going to be a folder for the storage of virtual disks VHD and VHDX format - which sort of Virtual Device gives the best performance IMG or IBV.

When using the SCVMM 2012 with the SMI-S agent is it best to create a new IBV or IMG file for each Virtual Machine and assign to the relevant cluster or host or create a VHD file on the Clustered volumes?

[Anatoly (staff)]

My question is if I shut-down the Hyper-V host would it wait till the writes had been flushed before shutting down?

In case of clean shut down - yes, definitely.

If the machine on which the Starwind target was shut down while a Hyper-V host has not flushed it's write cache, what happens to the unflushed write's still sitting in the Hyper-V's memory. Or is there some sort of communication between the Starwind target and the iSCSI clients that prevent the Starwind host machine shutting down and it waits till all iSCSI clients have flushed their caches?

I`ll ask you to clarify - you mean that the data was in the cache while SAN box hard failed? If yes, then the data will be lost. To prevent this - use HA;)

Should the server on which the Starwind software has been installed be dedicated or can other Windows file services DFS, DFS Replication run?

You can consolidate those ones - just make sure that there will be enough of hardware resources.

With regard to a dedicated server just running Starwind target software for iSCSI clients on a 1Gb network (my clients have usually tight budgets and 10Gb network gear is too expensive) how could you achieve the best network throughput:-
1). MPIO and add extra NICs to both the Starwind server and its Hyper-V and other iSCSI clients
2). Link aggregation with a managed switch and add extra NICs to both the Starwind server and its Hyper-V and other iSCSI clients
3). Combination of the above

I think it`ll be cheapedr to use 10Gigs starting from the moment when you`ll decide to use 3*1Gigs, so choose wisely. But anyway, if you`ll still decide to stick with couple of one gigs, I`d recommenr to use MPIO.

If it were not a dedicated server and other File services were running DFS, DFS Replication, SMB shares how would this affect the question above regarding maximising network throughput?

It all really depends on how actively those apps will be used.

With regard to HYPER-V clustered volumes which is primarily going to be a folder for the storage of virtual disks VHD and VHDX format - which sort of Virtual Device gives the best performance IMG or IBV.

Both evices can show great results, you just need to perofrm tests to verify it, and maybe tunning.

When using the SCVMM 2012 with the SMI-S agent is it best to create a new IBV or IMG file for each Virtual Machine and assign to the relevant cluster or host or create a VHD file on the Clustered volumes?

In my opininon, it`ll be better to store VMs on CSV, and not to use "target per device" policy.

[ticktok]

With regard to MPIO, I have been unable to connect stand alone server to a target that does not have clustering support set on it. What if any affect does turning clustering support on have on performance? Am I right that turning on clustering for a Starwind target is required to use Windows 2008 R2 MPIO on iSCSI devices?

[Anatoly (staff)]

Enabling clusterring support allows you to have more then one iSCSI connection to this target, thats it.
It will not decrease performance, moreover - it designed to increase it.