Automated Deployment to FC LUNs, Best Practice

[#1]

We are doing a deployment for VDI using approx ~ 40 FC LUNs and about 500 VMs spread across them on a FAS 3040 in multiple sites. Given we are using dedup, what is the best method for deployment?  We were wondering from a dedup prepective if it would be better to deploy approx 10-15 VM per LUN starting with LUN0. Basically, we wanted to deploy the VMs so that we get the best storage efficiency; we do not want to waste space by deploying the inital VMs across all the LUNs. The inital deployment will only be about 40 VMs per site. What kind of deployment would scenario would you recommend?  Can provide additional details if needed.

[#2]

David,

I just want to be sure I understand your situation. You need 40 VMs per site connected via FC and each site has a 3040 correct? I'm sure what you meant by the initial VMs, You are asking if you should fill up the first LUN before adding the second?

If that is the case, it doesn't really matter. Since Dedupe runs regularly it will dedupe as you add VMs so you will use the same amount of space if there is 3 VMs in the LUN as if you had 20 VMs in the LUN.

Some things you could consider to maximize your dedupe would include increasing the number of VMs per LUN. The official VMware best practice is 20-25 VMs per LUN but I just had a conf call with them and one of my VDI clients and we asked exactly how many we could safely place in a LUN. The answer was the classic IT "it depends". 20-25 is typical but if you have low utilization VMs or if you don't tend to use VMware snapshots, you can place more VMs in a LUN. 30-40 was a viable for my client. This will increase your dedupe ratio.

Another option is you could place more than one LUN per volume. Dedupe happens at the volume level so this can increase your dedupe as well. Doing this though can effect your restore capabilities if you plan to snapshot the VMs in the volume so you have to think that though first.

Your title says Automated Deployment, are you perhaps wondering if there is a tool to deploy the VMs remotely to each site?

Keith

[#3]

David,

Having multiple VMs in a LUN will provide you the best storage efficiency. As Keith mentioned, it depends on how many VMs you can have in a LUN. But 20-25 is a typical scenario. Also, to achieve higher storage efficiency you can have multiple LUNs in a volume.

We have a tool called Rapid Cloning Utility that can help you automate the provisioning of VMs in VMFS LUNs. I would highly recommend using it to deploy your envionment. For 250 VMs per site and 25 VMs per LUN and with one LUN per volume, you are looking at 10 volumes per site and each volume will initially have 80-90% storage efficiency. It leverages the NetApp cloning and storage efficiency capabilities. It can be downloaded here:

http://now.netapp.com/NOW/download/tools/rcu/

Configuring scheduled dedupe on volumes hosting LUNs will ensure maximum savings throughout the environment lifecycle as new writes happen.

Also, the section 4.15: "Deduplication and LUNs" in TR3505 provides guidance on deduplication best practices for environment with SAN LUNs.

http://media.netapp.com/documents/tr-3505.pdf

Regards,

Abhinav

[#4]

Basically the configuration is currently 8 250gb luns per volume, so 5 volumes. I was not aware of the ratio being that high, but these will be high utilization. So, based on your reply we could initially deploy 40  vms to the first volume as opposed to say 10 vms to vol 1-4. Any caveats to deploy one way vs the other, from a performance perspective. The concern came with the automation process actually needing the entire space when deployed. Since only volumes are deduped, would a multiple volume deployment have any advantages? I think we are talking over a short time frame of say 2-3 hours. I just saw a reply from Abhinav about RCU, I am downloading it now.

[#5]

Hi David,

Check out the following thread where I provided information on NetApp resources that can help you size for scalability.

http://communities.netapp.com/message/8451#8451

Hoe this helps.

Regards,

Abhinav