I have an issue that involves load-balancing VMs across multiple host with local storage (Fusion-IO).
We are using VMware View 5.0.1 on two sites with vSphere 5.0 (one cluster) and vSphere 5.0.1 (one cluster). Both clusters exist of 3 physical identical vSphere hosts and are using FusionIO drives as local storage (also FusionIO cards have the same specs).
This is what happens. We have created multpile pools (all linked clones), both Dedicated and Floating which are using all available hosts in the cluster. This means that all VMs that will be created by the pools are supposed to use all hosts and local datastores. This actually works fine for a few months now (btw performance is incredable because of FusionIO drives). We are in the middle of migrating our fat clients to VMware View. This week we noticed that there were some errors in View. This turned out to be some VMs that were not able to "Power On" because physical memory of ONE host was not available (98% used). The other hosts were only at 20% memory usage.
When we started to troubleshoot this issue we noticed that View Manager was trying to create the VMs on the vSphere Host that had the highest disk space available. This has to do with the Dedicated Pools (use more diskspace then Floating Pools) being not evenly divided between all de vSphere Hosts in the cluster. A way to fix this was to rebalance to Pool, however during office-hours we cannot do this.
I understand that this is "standard behaviour", but I was wondering if there is a possibility to change this? I do not want available disk capacity to influence the VM poplation in any way. In our case it would be far more efficient to use "Lowest Population Density". Local datastores running out of space will never be an issue in our current design. Even if this might happen we have our monitoring system to provide us with a warning. I want to divide the load of the VMs based on physical resources (cpu and memory of the vSphere Host).
The only information I found in the forums is the paragraph down here. This is exactly what I want but I have no idea how to do this?
If you choose multiple datastores for linked clone pool creation, View selects a datastore based on "Lowest Population Density" (and not based on largest available space) For example. If you have a 100 GB datastore with no VM inside, and 200GB datasore with 80 GB used space, though the 2nd datastore has 120 B free, View selects the first datastore, as it has lowest population density.
By doing this it ensures that a datastore which has less population is always selected and thus by balancing the storage usage
Does someone know a way to change this, if there even is?
Thank you in advance!!
I'm not aware of any option inside the View settings that allow you that forumla. Given the setup I would open a ticket with VMware support to see what they recommendation is. Please report back and lead the community know what their response is.
Also, it's nice to hear that the FusionIO stuff works so well. I saw them at VMworld and spoke to a couple of their team about VDI.
We have a similar setup here with 2 vSphere 5 clusters of 7 hosts in each using two 640GB MLC in each host and 3 Connection Brokers in each Cluster. We are only at the initial stages of ramping up to a large number of desktops.
We have a challenge in relation to load balancing and am very curious how you are managing yours? Our typical Users contsatly switch terminals (dur to work practice), so load balaning via source ip wouldn't necessarily guranatee their 'running' desktop as the end devices would be load balanced via source ip. Was wondering if you have come across anything like this before in your environment?
We are using a Cisco ACE load balancing configuration. Would love to have a quick conversation with you and maybe could assist each other in experiences to date? I can be emailed at butlpau (a-t) hotmail.com
Best of Luck,
I have opened a support ticket with VMware, this is there answer.
The response from engineering on this issue was not what I hoped for.......
Basically they changed the decision method for the VM allocation and never implemented the option to allow the admin to choose the method.....
The decision was made two years ago, so I have to assume it did not foresee the type of local storage coming on-stream.....
I have discussed the issue with the Senior View representative and she agreed that at this time, the best option would be for you to open a Feature Request for this.
Once you have the FR open, please ensure your TAM [technical account manager] or SE [sales engineer] promote this feature for you with PM [product management]....
There input will ensure your FR gets a higher profile and a better chance of being incorporated in a future release.....
My apologies that we do not have an immediate solution for you here...
I will open a Feature Request for our problem. In my opinion there are more customers going to experiencing this problem.
This article is about load balancing VMs created by VMware View across physical vSphere servers in a cluster which are using local storage (e.g. FusionIO), not for load balancing on network level. We are not using this in any way. Just one connection server protected by vSphere with DRS and HA.