FAS8020 cDOT8.3 and VMware

[machinist]

Hi all, I'm new to NetApp and have just installed a shiny new FAS8020 (switchless HA) with 3x DS2246 disk shelves

I have been reading through the latest version of the NetApp whitepaper, "VMware vSphere 5 on NetApp Clustered Data ONTAP - Best Practices", but it is not clear to me how to layout my SVMs and LUNs.

We will have all our VMware virtual machines .vmdks (around 20) on the NetApp as our host servers only have an SD card in them to boot ESXi from.

Do I need to create just one SVM, with one volume per virtual machine, and multiple LUNs within the volume for each disk I'll want to create for the virtual machine's O/S??

I hope I'm making sense!

Any advice greatly appreciated

[chott]

Hi,
My name is chriz, i'm a principle architect at NetApp. Congratulations for choosing cDOT and the 8020 🙂
Thank you for bringing up that question, i'll make the author aware of the missing clarity.

A single SVM should suffice, SVMs can be used for secure multi tenancy, meaning securely seperating different user groups / customers or applications on a single cluster.
If I understand your usecase correctly, none of those requirements do exist. Sometimes it can make sense to just create SVMs for logical, organisational purposes too.

As for the volumes:
I would recommend creating more than one volume, for your esx environment maybe 3-4, just to make sure you get IO requests parallelized accross all components of the system and stay flexible in terms of sizing / capacity for the future. As you are probably using dedup, you might want to think about grouping certain OS types within the same LUN / datastore which will then increase the probability of finding identical blocks in the OS.
With esx via a blockprotocol (iscsi / fcp) each lun will represent a datastore which will hold one or more vmdks. You might want to put vmdks of the OS disk of a VM in one datastore / Lun and the actual data disks into a different one. Also you can think about a tiering strategy or QoS Limits on the LUN level. CLuster Data ONTAP and esx are very dynamic and flexible, so you can always change things later.
If you where to use NFS instead of block, each Volume would represent a datastore, holding several vmdk files.

How are you planning to do your backups?
Hope this makes sense, let me know if i can add more to help.

If you think this is a usefull response please mark as answered or KUDO, to make it easier for others to find.

Cheers chriz

[machinist]

Hi chriz,

Thanks for your detailed reply!

I've created a single FlexVol SVM with Security Style "UNIX". 90% of the volumes I create will host CentOS-based VMs.

One thing that came up was whether I should create an SVM management LIF?? I chose not to, but the guidelines seem to think I'd need one for SAN protocols (I'm using FC). Can you advise on that at all?

Many thanks, Mark

[chott · Accepted solution]

Hi mark,

i'll do my best.

It is a Best Practice to create a Mgmt LIF, as it is used by SW like i.e. SnapDrive to help you manage your SVM and Data in it.

It also let's you seperate networks in case you have a data network (10G or FC) and a mgmt. network. This might come in handy in case you would like to delegate SVM mgmt. to an application owner and don't want him to have full access to you cluster (as you can also manage your SVM through the cluster mgmt lif)

Reasons for not creating a mgmt. lif could be a constrain on IP Adresses or LIFs.

hope that makes sense.

cheers chriz