deploy vm from template

[spinner]

hello,

deploying vm's from template's takes anywhere from one to two hours per vm

is there anyway to speed this up? the template is a compressed image, should i try an uncompressed one?

i am running dell poweredge 1950 III with 32GB of RAM, and dual quad core processors

the servers are hardly loaded and my storage environment is iscsi with wasabisx2000

the vm's are windows xp professional with sp3, the vmdk files are 20gb, but only 8gb is used

can anyone share their experience with respect to time it takes to reply vm from template?

i am using customization, but very basic, such as product key, ip, no scripts at all

thanks...

[weinstein5]

where are you templates stored? - are they stored on an NFS datastore? How is your networking configured?

[spinner]

the templates are stored on an iscsi wasabi 2000sx system

the lan has two cisco 3570 switches gigabit ports

the esx servers have 6 network cards, 2 for lan, 2 of iscsi, and 1 for service console

[vmmeup]

Even though there is only 8Gb in use it is deploying the full 20Gb on to the VMFS filesystem. Is your template on the same datastore that you are deploying to? Depending on how you have your storage carved up you might gain some performance by putting the template on a different datastore then what you are deploying. If you have have luns that you can share from different disk sets and you have the template on one and are deploying to another then you are seperating yoru reads and writes which should help out. Also ESX uses difference queues for each datastore which should also help. Are your NICS active actice and if so how is the network configured on the wasabi? You could also try NFS storage might give you some better performance. Are you running iSCSI NICs in the ESX Host?

[spinner]

i am using round robin, yes i am running iscsi nics in the esx host, i.e., nics with iscsi TOE

i am doing 6 convertions at a time, on the same datastore, perhaps that is my problem

i am going to try an uncompressed template, and will do 2 at a time on different lun's, let's see what happens

i have only 2 LUN's so having multiple datastores is pointless, because the datastores will eventually boil down to two LUN's on the wasabi

thanks

[vmmeup]

Do the luns consist of different disk groups? If you have 12 spindles and the 2 luns span them all then you won't see much benefit either way. If you have 2 luns and they each have 6 spindles then you should see a performance increase. If you were to deploy two to one lun and two to anothe then they would be writting to different disk groups where one big disk group they would all be writting to the same disks regardless of how the luns are broken down.