2010-01-28

VMFS or NFS

We are outgrowing our current storage capabilities, and virtualization is one of main "customers" that is causing this.

So we are now looking for a 2nd storage that will enable us to continue on our journey forward.

I am a great fan of NFS for usage as a datastore for ESX. There are numerous reasons:

  • Backups can be done on the storage level with little or no impact on the Host.
  • In our specific environment, the backup speed we are getting with NDMP are excellent.
  • The ability to mount the file systems elsewhere is extremely beneficial

This is just to name a few.

But NFS has its downsides. ESX does support NFS as your storage platform, but I do not think that VMware are completely "gung-ho" about using NFS as storage.

I do think that the future will be be more in VMFS - it is a proprietary File System that is VMware's. they know it inside out, and have been tweaking it more and more over the years, to provide better performance, to be more stable and so on… NFS on the other hand is not something that VMware will have any much of a say in how to tweak or improve the protocol or its performance.

A small example. Disk metrics. How many of us know how to measure disk metrics for NFS datastores, in the vSphere client? Metrics Per host? Metrics per VM? Same question I can ask about the metrics in ESXTOP. They are not there. Disk metrics are for VMFS volumes only, which means iSCSI / SAN.

True you can get the metrics out of vscsistats but it is not an intuitive tool, and not the easiest to figure out. What you will get from the GUI is network throughput for the whole VMkernel but you will not be able to tell which VM is using the disk throughput.

From performance papers that VMware have published, performance of FC is always the best on all counts, NFS and iSCSI are almost the same.

So which to go for? The answer for that one I think would be go for tiers.

Not all your dev and test machines will not need the speed and redundancy that your will get from multiple HBA's, on your array that has a gazillion SAS 15,000 RPM drives. They most probably will be able to get by with a lot less than that.

I have learned a lot of the past few weeks regarding the storage planning and what has to be taken into account. The more I learn, the more I come to realization (and I am trying to teach others)

SIZE IS NOT EVERYTHING!

Ok maybe a bit sneaky of me, let me rephrase - the amount of TB is not the most important thing, the important thing - is can your storage design provide the amount of IO that your Virtual Machines will need.

Seeing that we will be going for more critical Apps on our Virtual Infrastructure - the more we will be needing faster disks, and ways to get to them, it looks like SAN for me.

Now to solve the small problem of how to backup all of these VM's properly :)