Question:
When I create a linked clone desktop pool on local storage, a replica is created on each local datastore where linked clones are created. When the parent-VM is located on shared storage, accessible to all hosts, how is that parent VM copied to the local datastores? Is that over the management network interface or over the storage interface? And is it possible/supported to have the parent VM not on shared storage, but on one of the local datastores?
I think it is copied over the network. Why do you want to have the Parent VM on the local datastore? If you want to update the your environment and for some reason Host where Parent VM resides crashes it will take some time for you to recover. Not a good situation.
I think it may be over the network too. You could easily test this by performing the copy and see if you see the spikes on the storage or the network adapter. The parent VM can live wherever you want it but like the post above stated your creating a single point of failure.
That doesn't mean your clones would stop working because they are linked to the replica but without backups you would have to possibly rebuilt your parent VM and then point your pools back to the new parent VM.
I don't want to have the parent VM on local storage, but that's something I ran into at a customer and I wondered if that was supported. I was under the impression that the parent VM needed to be on shared storage so it could be copied to a replica on each local storage. But that doesn't matter if it is copied over network, it could be anywhere. The VDI-hosts would only need to have local storage and no shared storage whatsoever.
When deploying a new pool or when updating a pool, copying over network is very slow. It would be faster using the shared storage.
There are lot of factors that decide the pool deployment time frame. Having SSD drivers with deduce, using QuickPrep and not Sysprep, customization scripts if any... all add or reduce the time it takes for the pool to be created. By design initial time process wherein Replica is created is slow because of copying process.
What I've noticed:
The master image is registered to a host in the cluster and is located on shared storage. When a pool is deployed to the local datastores on the hosts in the cluster, the master image is copied to a replica on all local datastores. The copy to the local datastore on the host where the master image is registered is fast because it doesn't need to go through network: Shared storage -> local datastore. Copying to any other host in the cluster is slow, because the copy is from the host where the master image is registered on, over the management network: Shared datastore source host -> Management Network source host -> Management Network destination host -> local datastore destionation host.
Why not:
Shared datastore source host -> Local datastore destionation host.
This could be done if vCenter was smart enough to know that the destination hosts have direct access to the master image (which is located on shared storage).
When not using local storage but only shared storage for the replica's (and link clones), this is not an issue because the host where the master image is registered on takes care of copying the master image to all the required replicas.
I have come across the same issue.
This KB article roughly describes what is happening here: http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=208481...
Has anyone come across a workaround? The clone to where the source VM (20GB in size) is registered copies in 5 to 7 minutes, however I am experiencing a clone time of up to 8 hours to the other hosts!. The source VM is located on a dedicated 6 x SSD RAID storage shared by iSCSI and the local datastores are write intensive SSDs (Dell R730 servers, 48 HT cores, 386GB RAM, 12G SAS3 backplane, H710 RAID controller). The fact is clones to the mounted host in less then 8 minutes consistently proves it's not a storage performance issue. I have tried mounting it to all 3 of my VDI hosts, it's always fast to the mounted host and very slow to the rest. I am currently running dedicated 1Gig switching for this environment, going to 10Gig switching as soon as the switches arrive, but the current ports are not reaching capacity.
Anyone come close to solving this problem, or have any workarounds?
I am also having the same issues with deploying virtual machines from shared storage to a number of hosts running local SSDs in a cluster.
The attached virtual machine to an selected host in the cluster will deploy in minutes but the other hosts take up to an hour plus.
This KB article roughly describes what is happening here:
I have tried the KB 2084819 which does specify ESXi 5.1 update 3, I am currently running ESXi 5.5 and vcenter 5.5 update 2.
My storage network is 10GB, vMotion Network is 1GB and VM network is 1GB.
Any ideas how I can fix the issue, one idea is to upgrade the VM network to 10GB but this did work great with ESXi 5.0.
Did you change these parameters Mentioned in the Article VMware KB: Cloning or deploying from a template takes longer after upgrading to vSphere 5.1 Update 2...
n VMware Infrastructure Client (VI Client),
In vSphere Client and vSphere Web Client,
Thank you for your response.
Yes, I have actually tried it on both of our VDI deployments sites with no luck at all.
I also have rebooted the vCenter Servers etc and still all traffic is routed through the VM network.
As I mention before 5.1 everything was perfect.
what is the exact version of Vcenter 5.5 and Esxi
http://kb.vmware.com/kb/1014508
We have installed release ESXi 5.5 Express Patch 5 2014-12-02 2302651 and vCenter Server 5.5 Update 2 2014-09-09 2001466.
Thanks
Hello David,
Apologies for the delay in replying because of the Holiday Season.I think you are hitting a known issue and wanted to check if you have a ticket open with VMware
I believe
From shared-storage to shared-storage, it appears both reads and writes appear on the source Host.
From shared-storage to local-storage, it appears reads occur on the source Host, write on the destination using NFC.
Could you please Check if The Source has access to the datastores .I know it might not be a good workaround Could you perform the same step on the shared Storage if possible.
Hi and thank you again.
I have not opened a ticket with VMware at this stage.
Yes all my hosts have access have to the shared storage, I have attached a diagram of our networking model, which is standard with all storage running on 10GB and VM network on 1GB etc.
All our servers are DELL R720s and the storage is DELL Equallogic and DELL PowerVault.
From shared-storage to shared-storage, it appears both reads and writes appear on the source Host. Yes 100%
From shared-storage to local-storage, it appears reads occur on the source Host, write on the destination using NFC. again 100% and this the issue.
Cheers
David.
Hello David,
Do you mind Helping me with the SR# please