v8 vSphere Final Data corruption when using FLAT

[haslund]

I have been testing the new final version and seen some of the issues I saw during RC (which was fixed) come back.

This setup is running:
Starwind 8.0.6784
All LUN are created with 512 MB L1 write-back cache.
vSphere 5.5 build 1623386
VM running Windows Server 2003 with SP2 slipstreamed - fully unattended. Tested OK on local HDD and SAN several times without any issues.

First test, FLAT 10 GB iSCSI LUN upgraded from 8.0 RC:


Second test, FLAT 10 GB iSCSI LUN freshly created:


Interestingly this does not happen on a freshly created LSFS iSCSI lun.

Any ideas?

Feel free to let me know if you would like a remote access to test for yourself.

[anton (staff)]

We'll take a look for sure. Do you happen to have a matching StarWind log for these cases?

[haslund]

I know Bohdan did many remote sessions and tests to look into this.
Did you guys figure out something?

[anton (staff)]

Yes we did. You config (many hypervisors hosted inside each other, StarWind runs as a VM on vSphere also virtualized and feeds storage to outside the cluster etc) is pure academic one and is not supported in a production. No single paying or playing customer from around 10,000+ we have ATM has anything like that I'd suggest you to simplify the setup as I was told MSFT target provides the same bad results (no StarWind code at all) which makes us think it's not really our issue.

I know Bohdan did many remote sessions and tests to look into this.
Did you guys figure out something?

[Bohdan (staff)]

StarWind-VM feeds the storage for the pair of nested ESXi hosts.
Yes, this is a non-production scenario.

Update from me.
I have the installation ISO.
As soon as it:
1) happens for MS iSCSI target too and
2) does not happen for non-nested hypervisor environments;
we are planing to research such scenario but a little later, after we finish the tasks with a higher priority.
I believe we'll be able to check it at the end of this week.

[haslund]

After further digging it appears this issue is down to misconfiguration from my side - somewhat at least.

For my test scenario I have been using Windows Server 2003 SP2.
Starwind iSCSI was configured to use a sector size of 4096.

However as it turns out Windows Server 2003 does NOT support a sector size other than native 512 bytes (512n).
Microsoft has a table available that shows this @ http://msdn.microsoft.com/en-us/library ... s.85).aspx

From the Microsoft website:

Windows XP, Windows Server 2003, and Windows Server 2003 R2 do not support 512e or 4Kn media.
While the system may boot up and be able to operate minimally, there may be unknown scenarios of functionality issues, data loss, or sub-optimal performance.

I would like to thank the Starwind team for their deep cooperation on troubleshooting this issue - world class service which I have been very! impressed with.
Most other vendors would have given up much faster.

[anton (staff)]

Thank you for your update! Nice to know it was easy fix