VMFS Sizing, Snapshots, Metadata.

[conyards]

So I've got a pretty well bedded in infrastructure, running some 2000 Production VMs, 100 Hosts, 318 VLANs and 975 LUNs across 21 clusters.

VMs run across multiple LUNs either configured to host single or mulitple VMs dependant on purpose (OS, PF, DS) and SAN replication status. Sizing of those LUNS has always been based upon number of VMs * VMDK size + VSWAP Size + 10% overhead for metadata (which may be a little large, but having had to support the infrastructure when it wasn't clearing the journal and LUNs ran out of space, I would much rather be safe than sorry).

This estate is in the process of being upgraded, and to this end I'm taking a close look at VDR to backup a portion of the estate. This will obviously require VMs to be relocated to larger LUNs as snapshot space is not factored in. I will be looking to use the formula below to provide sizing recomendations for the OS LUNs and would appreciate some feedback as to what other people are doing, be it a formula or a finger in the air

The below is just for a LUN housing OS' (VMX files) as this is the default location for snapshots, for this reason the formula uses data about VMDKs housed on other LUNs.

The formula assumes a maximum snapshot size of 15% of the largest VMDK that makes up the machines hosted from the OS LUN, however this figure can of course be factored up for VMs with a higher rate of change.

((Number of VMDKs on the LUN * Largest VMDK on the LUN)(VSWAP*Number of resident VMs))((Largest VMDK on the hosted VMs * number of VMs)/100*15)

The resultant number that comes out of the above has a further addition of 10% for Metadata etc, and is then rounded up.

Thoughts? How do you derive LUN sizing?

Simon

[AndreTheGiant]

Have a look also at this document:

Andre

[conyards]

Hi Andre, not sure what vSwitch and NIC white papers have to do with VMFS sizing...

[AndreTheGiant]

Sorry... wrong document...

The right is:

Andre

[azn2kew]

Looks like you have large enterprise infrastructure and sounds like a performance hit if you do not plan your LUN I/O correctly. I'm not sure what types of SAN you're using but each SAN has different recommendations in sizing and limitations. For flexibility, NetApp NFS should be good since you can extend the LUN on the fly. The management snapmirror and SMVI tools should be great for your backup to DR site or for archive but again not sure types of SAN you got. In order to migrate and upgrade seamlessly, you may need to introduce new SAN for staging purpose and carefully migrate them using storage vmotion, or LUN replication or whatever feels right for you.

I have the impression you have multiple types of SAN environment and for high load systems you're using FC and others may be fine with iSCSI/NFS solutions. The formula you have is kinda complex but it should have enough space but if you can utilize thin provisioning and deduplication solutions then you don't have to worry much about over allocating those LUNs. Interest to know types of SAN and how you architect it for your environment.

If you found this information useful, please consider awarding points for "Correct" or "Helpful". Thanks!!!

Regards,

Stefan Nguyen

VMware vExpert 2009

iGeek Systems Inc.

VMware, Citrix, Microsoft Consultant

[azn2kew]

Forgot to comment about your VDR backup solution, isn't it VDR limited on number of virtual machines can be backup per appliance? If you have to backup that much of VMs, you may have to deploy multiple VDR appliances. For backup purpose, I would look at SAN replication or snapshot so you can restore quickly and register the .vmx file and have them online quickly.

But if you require to backup all those VMs, then I suggest using more enterprise solution like Vizioncore vRanger, Veeam Backup, PhD esXpress as well. They may be pricy for your 2000 machines

If you found this information useful, please consider awarding points for "Correct" or "Helpful". Thanks!!!

Regards,

Stefan Nguyen

VMware vExpert 2009

iGeek Systems Inc.

VMware, Citrix, Microsoft Consultant

[conyards · Accepted solution]

Hi Stefan,

Yes you are correct there are several different SAN types in the infrastructure, supplying fibre attached differing disk types. the majorioty of this disk is virtulised behind either an SVC or an USPv.

For clarity I'd only be looking to backup a smaller subset of VMs using the VDR solution, say for example the 120 environment management VMs.

I rekon whilst the equation I've stated above isn't perfect, it does fit for the infrastructure I'm dealing with.

Thanks

Simon