XenServer 5.6 + StarWind HA problems

[xonus]

We're trying to connect XenServer 5.6 to Starwind (5.4), and are having problems getting it to work.

We have the following setup:
[SANbox1]
192.168.1.215 - management
10.10.1.1 - SAN link #1
10.10.2.1 - SAN link #2
10.10.0.1 - 10Gb crossover link to SANbox2
[SANbox2]
192.168.1.216 - management
10.10.1.2 - SAN link #1
10.10.2.2 - SAN link #2
10.10.0.2 - 10Gb crossover link to SANbox1
[CitrixBox]
192.168.1.213 - LAN traffic
10.10.1.13 - SAN link #1
10.10.2.13 - SAN link #2

We've set up the 2 SW servers, and created a high availabilty target. I can connect from my desktop (w7) to one of the SAN boxes no problem, and everything works fine. We've switched off the firewalls on both SAN boxes (they're running 2008R2 DC). We also did the iScsiDiscoveryListInterfaces changes as outlined in the XenServer pdf.

However, as soon as I connect from XenServer into SANbox1 it says its partner is un-synchronized almost immediatly. If I run a fast or full resync, it'll show as normal until I write something on the SAN, whereafter it shows un-synchronized again almost immediatly.

This is without multipath enabled in XenServer. If I enable multipath in the GUI and try to connect to the SAN boxes, I get an error while trying to format and it drops the connection again. I can only get the connection to work with multipath off - I can switch it on after its connected.

We've tried it both on a freshly installed XenServer, and a XenServer where I did the recommendations outlined in this forum post http://www.starwindsoftware.com/forums/ ... t2053.html

We also tried disabling the direct link between the SANboxes - but no change. I've searched and searched, but haven't found any solutions that works. We're at our wits end here, since we simply cannot get Starwind and XenServer 5.6 to play nice together.

It seems like its a config issue in XenServer (since it works from Windows), but so far we've been unable to locate what the problem is. No matter what we do, the StarWind shows as unsync'ed within a few seconds.

[Max (staff)]

Hello Xonus,
Configuring Xen to work with HA is a bit more tricky
You should use this document for reference:
http://www.starwindsoftware.com/tmplink ... Guide.docx
Please feel free to ask any questions if something is unclear.
BTW: Do not touch the Sync channel if you want to keep your data safe

[xonus]

Thanks Max, I'll give it a whack (before bringing out the hammer and REALLY giving it a whack). I'll let you know how it goes.

I know - don't touch the cable. Its just a test-setup for now - we'll set it up a bit more stable in production.

[Max (staff)]

You're welcome, I'll put a detalied PDF on our website in the nearest future.
Yeah, the cable is kind of a Red button, a small thing which can destroy everything if treated incorrectly

[xonus]

I've tried the document - which was the same as on a forum post, but it didn't change anything. As soon as we connect to SW the data between the two SANboxes becomes un-sync'ed. I can resync it, but that only lasts until I start writing new data to the SAN.

We don't really know where to go from here, as nothing we've done so far has made any difference. We continuously end up with the primary SAN saying the secondary SAN is unsync'ed.

[Max (staff)]

Desynchronization while connecting one of the targets is a normal thing for Xen. Please ensure that you've connected both targets to Xen server, also pay attention that the unsynchronized target will not accept client connections until it's synchronized. So this should be done this way:
1. connect first target
2. resynchonize (if needed)
3. connect second target

[xonus]

We have 4 paths (2 sessions) active to the two SANS. After we connect, we've done a resync (it can't do fast, so we've done a full). But the unsync'ed problem keep happening. As soon as we start writing data to the SAN it goes into unsync'ed again.

[Constantin (staff)]

Unfortonatelly the reason of such behavior is that Xen send 512kb block instead of 256kb blocks. You can resolve this issue in following way: go to

Code: Select all

HKEY_LOCAL_MACHINE\SYSTEM\ControlSet001\Control\Class\{4D36E97B-E325-11CE-BFC1-08002BE10318}\0001\Parameters

and change values of paramets MaxBurstLength and MaxTransferLength from 4000 to 8000

[xonus]

That did indeed do the trick. Thanks Constantin.

You should write this into the XenServer docs - I haven't seen anything about this anywhere.

[xonus]

And btw you're missing a 0...

[Constantin (staff)]

Oops! We`ll add it there!
This trick should work, I have tested it by myself! I suggest you to try to reboot StarWind box.

[Max (staff)]

I'd like to note that these values do not work for Xen 5.6 fp1

[basscom]

Hi Max,

What values should we use for XenServer 5.6 FP1? I'm currently experiencing the same problem with a RAID1 Mirror configuration on StarWind. Mirror goes out of sync as soon as data is written to the SAN.

Chris

[michal]

Hi Max,

What values should we use for XenServer 5.6 FP1? I'm currently experiencing the same problem with a RAID1 Mirror configuration on StarWind. Mirror goes out of sync as soon as data is written to the SAN.

Chris

Here's the thread with the working settings:

http://www.starwindsoftware.com/forums/ ... t2383.html

[basscom]

These settings have been tried after being referred to the thread you linked from this thread:

http://www.starwindsoftware.com/forums/ ... t2376.html

It is also a slightly different scenario since I'm not running HA, only Mirroring. Perhaps it should have worked but I've tested with 2 different XenServers now with the same result: mirrored volumes become unsynchronized when data is written across the LUN from XenServer connections.