zenomorph
Contributor
Contributor

Storage design for File Server VMs

Jump to solution

Hi there,

I am in need of some assistance. We are planning the SAN for our File Server VMs which currently have volume sizes of 2-3TBs this is because its for a single department and we cannot seperate the volume directories.

What we are using is CX3-80 with 300GB disks, what I was thinking was creating metaluns across 3-4 DAEs and presenting the LUNs to the ESX. However I'm not sure what LUN size I should use and VMware block size I should use if I'm going to have volume sizes of 2-3TBs. What I'm doing is consolidating some MS Cluster File Servers to VM's. Right now the File Servers have 4 1.6TB volumes and I need to migrate these to VM. So what I might have is 3-4 VMs attached with d:\ data drives of 1.6TBs each.

Can anyone give me some advice on LUNs sizing and VMFS sizing.

Many thanks

0 Kudos
1 Solution

Accepted Solutions
szelinsky
Enthusiast
Enthusiast

If you want to go with vmfs (2 TB if that is the size you need), block size will need to be 8 if you want your vmdk's to be 2TB. There will be more chance for wasted space.

I would agree with Fred and present the storage as RDMs.

View solution in original post

0 Kudos
6 Replies
FredPeterson
Expert
Expert

What is the actual usage of these shares? Are you monitoring the growth with something like Cacti or some other SNMP tool? You say you have 4 x 1.6TB right now, but how much is actively used on each? What has growth looked like?

I would probably just go with 4 x 2TB LUN's and present them as RDM's unless you still plan to do clustering. Also, what do the hosts utilizations look like? Could you consolidate down to less servers? You would present each RDM to the VM guest as a separate SCSI adapter to ensure performance of each is separated from the other.

zenomorph
Contributor
Contributor

Fred,

Thanks for your reply. The shares are actually just purely for file sharing and storage of typical MSOffice documents. Right now of the 4*1.6TBs probabaly each is about 60-80% used.

The utilization isn't that high so we could actually consolidate to less servers eg. from the current 4 Servers we can potentially consolidate to 2 VMs each carrying 2 volumes each of 1.6TB.

If possible I'd prefer to use VMFS rather than RDM because I'd like maintain the ability to do things like VMotion etc. such that we can mainain the HA/redundency ability.

Cheers........

0 Kudos
FredPeterson
Expert
Expert

Unless you intend to use MSCS across boxes you can still vMotion with RDM's - the LUN ID needs to be consistent across the hosts though I think if its in physical mode.

Can't remember my RDM and vMotion stuff at the moment other then it is possible.

zenomorph
Contributor
Contributor

Guys actually I prefer to use VMDK since it'll make things easier but what I'm not sure is how to parition the VMFS and the SAN storage. How big should I make the CX3-80 RAID groups and then the LUN sizes and the VMFS paritions.

Can you give any suggestions.

Cheers

0 Kudos
FredPeterson
Expert
Expert

I would view it no differently then if it were a physical host since in this case (due to the sizes involved) its not like you're going to plan to share the VMFS LUN with more then one guest and host simultaneously. You just happen to be using a shared storage volume for the sake of simplicity.

Myself personally I never want to share disk groups and spindles across VMFS volumes. Thats just asking for trouble when one LUN access starts effecting others. I have two ATA LUN's that like to do that. They share spindles and every once in awhile one of the LUN's decides to play bad and not share with the other children and it effects the second LUN.

0 Kudos
szelinsky
Enthusiast
Enthusiast

If you want to go with vmfs (2 TB if that is the size you need), block size will need to be 8 if you want your vmdk's to be 2TB. There will be more chance for wasted space.

I would agree with Fred and present the storage as RDMs.

0 Kudos