NFS vs iSCSI for virtual machines

steinooo

steinooo

steinooo

Soldato
Joined
31 Dec 2003
Posts
4,749
Location
Stoke on Trent
Hi all,

Could anyone advise on the above in laymans terms please?

I heard that it's better to store VMs on NFS rather than iSCSI, but I can't get my head around what the whole thing means, to me I thought NFS was a file system, like the kind you may see on a NAS etc, and iSCSI a transport mechanism to deliver data from the SAN to the server.

I'd appreciate a helping hand please!
 
You're right with the difference between NFS and iSCSI.

With VMware (and pretty much any x86 virtualisation) the VMs are just a few files (usually config files and the virtual disk file) so using NFS to store them is the same as storing any other files.

NFS storage can be cheaper when it comes to setting up shared storage for a virtual environment than iSCSI and it's almost always cheaper than a FC SAN but it does have some performance limitations.

NetApp have a couple of good documents on their website that explain the pros and cons of NFS, iSCSI, FC and FCoE for VMware.
 
NFS and iSCSI are file serving protocols, but you are kinda right about NFS being a file system - just a distributed one in essence.

It all depends on what version of NFS you want to use; NFSv3 from a good NAS filer to VMWare ESX servers is like shizer off a shovel fast (i know, ive set em up numerous amounts of times).
 
SMN said:
It all depends on what version of NFS you want to use; NFSv3 from a good NAS filer to VMWare ESX servers is like shizer off a shovel fast (i know, ive set em up numerous amounts of times).
Click to expand...

How would using NFSv3 differ from using the filesystem on a NetApp san, WAFL?
 
The last time I went to see them, Netapp recommended NFS over iSCSI quite adamantly (not an official line as I gather, just their strong unofficial recommendation). Then again they recommended FC over just about everything else too unsurprisingly.

You are right that NFS is technically served by a NAS (being a file level protocol) and iSCSI is served by a SAN (being block level).
 
So if I was just to put the VMs on the plain Netapp file system WAFL, and not NFS - do you mean to say that it won't be as fast?
 
Thanks guys, some good articles to read here.

I see it has some guidelines about LUN best practises in there which I need to read about, that's another concept I don't quite get yet.
 
Having recently upgraded my NetApp FAS from 270 to 2040 and getting an NFS license included I really wouldn't look back. ISCSI luns for VM's are bloaty and a management nightmare.

I had been running one lun per VM in order for me to back up and run in sperate storage space. This was I believe the best practices when first implementing the ESX architecture a couple of years ago.

So now with the NetApp upgrade and I just moved over to vSphere from ESX3.5 I have storage vmotioned most of my VM's onto NFS provisoned storage runnning 10-15 VM's in each NFS volume and it just makes my life a breeze.

Couple this with thin provisoning and dedupe on the NetApp vols my storageoverhead has decreased dramatically. My filer dispite the head upgrade was running about 92% capacity which however good the head affects performance, this has been reduced to well below 80% netapp bp's.

In short NFS makes management so much easier, grow VM datastores, dedupe...its great and no noticeable performance difference, although i have to admit i have no real metrics.
 
hello :)

I'm following this as I'm also in the same boat. I had stupidly (it seems) assumed that iscsi was the preferred/faster etc solution. So its interesting.
Any specific export options to increase performance or defaults are fine?
 
You must log in or register to reply here.
Back
Top Bottom