View 5.1 rebalance - what a mess

[gmtx]

About an hour ago I tried to rebalance a linked clone pool of 45 VMs on two esxi 5 hosts (latest updates and patches). I'm running the latest version of View. Each host has local SSDs that store the linked clones, and in today's rebalance I started with 18 VMs on one host and 27 on the other. I had two user sessions in progress when I started the rebalance, one connected and one disconnected. The balance of the VMs had a status of "Available". I configured the rebalance to wait for logoff for logged-in VMs and kicked it off. The rebalance started and I watched in vCenter as it shut down, reconfigured, copied, restarted, etc. a whole bunch of the VMs. When it was all through I had a mess.

Four of the VMs show as "Available" - all on the same host they started on. 32 of the VMs have a status of "Available (task halted)", seven of the VMs have a status of "Error - Configuration failure: Invalid configuration for device '0'", and the remaining two - the VMs with logged in users - have a status of "Connected (task halted)" and "Disconnected (task halted)", for the connected and disconnected VMs respectively. I have the same number of VMs on each host that I staretd with. In other words, it's still unbalanced.

This is the second time I've tried a rebalance operation over the last two weekends. Last week when I tired it I had similar results, but I thought maybe it was something I had done wroing.

Anyone know why this might be failing? Does rebalance need the VMs to be on shared storage to work?

Thanks,

Geoff

[Linjo · Accepted solution]

I would say it does require share storage since the rebalance is not taking esx-hosts into consideration only LUN:s.

// Linjo

[GeoffTX]

One other strange thing... I deleted and recreated the pool and View insists on putting 30 of the VMs on one host and 15 on another. I even tried it twice with the same results. Other than a few template VMs on each host (powered off), there is nothing else running when the pool creation starts, so I would expect View to balance the creation of VMs pretty equally across both hosts. Both hosts are identically configured with procs, storage, etc.

Any reason you can think of why it would create a 2:1 ratio? Maybe I should increase the pool size to 60 and see if it's better at math with evenly divisible numbers?

Geoff

[Linjo]

Again, this assumes shared storage.

If you want to use local disk you need to have one pool per host.

// Linjo

[gmtx]

OK, I should have read the docs more closely. They talk about load balancing and the impact of using local disk. (Basically, you can't do it.)

http://pubs.vmware.com/view-51/index.jsp?topic=%2Fcom.vmware.view.administration.doc%2FGUID-FB1CA0DA...

I'm still curious about how it's intially deploying to the two local stores - think I'll play with it some more tonight and see if I can make sense of the decisions it's making. From another post it appears to use "Lowest Population Density" as one of the deciding factors. I bet my powered-off templates are influencing that choice.

Geoff

[m-nara]

Hi Linjo,

At VMworld 2012 EUC2012 session, I heared the Hertz  Europe is using with link clone desktops with local storage .

It looks like multiple hosts which have local FusionIO in a single pool.

"one pool per host" is ture ?

I think if "one pool per host" is required, users must be entitled to multiple pools against a host failure. It is not convenient.

regards,