vMotion and XenDesktop disconnection messages

[khughes]

Hopefully this is the right area for this question, the boards have been broken up so much compared to the last time I was here.  I'm running into an issue now with vMotion not necessarily completing successfully but with causing a disruption in the network where clients are having their XenDesktop sessions pop up a message that they have been disconnected from the XenDesktop servers. This is happening not only when I'm moving the actual XenDesktop virtual machines but also servers on the hosts involved.  The servers have no issues with the vMotion what so ever.  I'm all but sure that Citrix will point to VMware as the issue and vise versa.

Our current setup is we have ESXi 5.0 U1 hosts, each has 3 vSwitches configured as follows- 3 pNICs go to MGMT / VM Traffic, 2 pNICs to vMotion (VMKernal), 2 pNICs to DMZ.  The pNICs for MGMT/VM Traffic and vMotion plug into the same switch stack of Cisco 3750x switches with no vLan's or tagging, and the DMZ is its own physical switch isolated.  I would think that dedicated pNICs for VM traffic and vMotion traffic on gigabit connections would suffice, and I would assume that the Cisco 3750X switch stack should be able to handle that kind of bandwidth without any issue.

I've been thinking about isolating the vMotion traffic onto it's own vLan but I'm not sure if that would do anything as it would still be plugged into the same switch stack.  Has anyone heard of issues with XenDesktop and vMotion before or where I should start to look for bottleneck issues? 

[TomHowarth]

I take it that the traffic is separated across NICs but you have a flat VLAN.  moving the traffic into their own VLANs could help if you placed some QoS on there. this would prioritize your VM traffic over your vMotion traffic, thereby stopping it flooding your switch and potentially causing a blip in your VM traffic between your DCC's and your XD Desktops

[Troy_Clavell]

are you by chance using provisioning services for your XEN environment?  We've had some inconsistencies with vMotion and XEN Desktops. Our environment is set to fully automated for DRS, but we don't have a lot of DRS events.  We haven't had an issue with the XEN desktops go offline, but there is a "pause".  That is one of the downfalls with VDI... any inconsistency in the environment will be seen immediately because you have users working on the VDI VMs all day long. 

I don't know what the exact fix is, but i'll see what our Citrix folks have done to try and reduce the timeouts.

[khughes]

Troy, yes we are using provisioning services as well, two balancing provisioning servers to be exact.  We did have an issue with those servers earlier in the month where Veeam backups were causing one of them to go offline until we restarted the stream service but that has been fixed.  Our users aren't going offline, but they get their screen grayed out and says the session is disconnected.  I've thought about extending the timeout timer so that while the sessions may pause for 10 seconds it wont time them out.   Thanks for looking into it Troy, also glad to see you haven't burnt yourself out of the forums yet!

Tom, that was my thought as well.  I suppose I should go to the switch stack to see if there is any bandwidth issues on the switch while vMotion is going on.  In theory it would be a good excuse to separate the traffic now anyways to move to a more acceptable networking best practice.