I’m a fan of using NFS for VMware; I’ve mentioned it before. I’m not the only one, either; there have been a number of recent blog entries from various people regarding the use of NFS for VMware:
Virtual Optics: Why VMware over NetApp NFS
Storage Nuts & Bolts: VMware over NetApp NFS: A Customer’s Testimonial
It’s great that this is receiving more attention in the spotlight, but I still have one question: where are the statistics proving NFS’ value in VMware deployments?
If NFS is equally as good as Fibre Channel or iSCSI—and personally I agree with Nick that most deployments would be hard-pressed to tell the difference—then where are the stats that show this? Or is it impossible to demonstrate that a VMware deployment on NFS is “just as good” as one on Fibre Channel or iSCSI? Does the value of NFS come in subjective measurements that can’t be quantified, and perhaps that’s why we haven’t seen any hard proof of the value of NFS when compared to Fibre Channel or iSCSI?
If you have some insight, please share it in the comments. I’d love to hear everyone’s thoughts on the matter.
9 comments
Comments feed for this article
Trackback link
http://blog.scottlowe.org/2008/01/14/proving-vmware-over-nfs/trackback/
Monday, January 14, 2008 at 11:20 am
Dan
Who’s On First: NFS or FC
Proving that NFS is better than FC is like proving that Fords are better than Chevys.
Both can provide enough speed to meet the 75MPH speed limit and both can safely get you work each day…
I have stats showing that NFS is just as good as FC for the way we drive, however your roads and loads may be much different than ours.
Soon there will be several other choices such as Infiniband, FCoE, NFS-RDMA, pNFS, and 10Gbit ISCSI all which bring options which will allow the flexibility and horsepower to virtualize all servers.
Maybe someone should create a Performance VM that runs a plethora of applications and spits out response times… Then we would be able to loadup the VM on different datastores for quantifiable performance results.
dp
Monday, January 14, 2008 at 11:45 am
pkillur
N.F.S. S stands for SOOOO SLOW when poorly implemented, which is how it normally is
-From personal experience working at VMW GSS.
Monday, January 14, 2008 at 6:41 pm
Nick Triantos
Scott,
You need to keep in mind that the first protocol VMware supported years back was FC. So, naturally, people would tend to gravitate towards it.
Now, I’ll take you back in the mid to late 90s. At the time NetApp was out on the street evangelizing the virtues of NAS and the fact that people can safely run Databases on NAS. The competition at the time vehemently oppose it. Then Oracle comes out supporting NAS, IBM with DB2, Informix, Sybase. Then all of a sudden we started to see whitepapers and architectural blueprints from those who opposed it!!!! History repeats itself.
Server virtualization enables a grid type of architecture. If you visit HPC environments, the protocol of choice isn’t Fibre Channel. It’s NFS. Has anyone ever wondered why?
Server virtualization is disruptive technology. This is a compliment. But it also disrupts/breaks IT processes and procedures. Backup’s an example of that. Furthermore, it requires that larger ITs re-think their architectures, as deployments and management based on the traditional model ultimately leads to a collision against a brick wall.
Cheers
Monday, January 14, 2008 at 8:49 pm
slowe
Dan,
I don’t want to prove that NFS is better than FC, just that NFS is an equally valid choice for VMware deployments. I think that both FC and NFS (and iSCSI) each have their strengths and their weaknesses and their appropriate places and situations for deployment. What needs to happen is that the stats that work for you need to be formalized and documented so that others can see where NFS fits and doesn’t fit. As it is, people are just guessing. (More on that below.) Thanks!
Nick,
Oh, you’re absolutely right, no question about it. VMware’s early support for FC was what catapulted it into 1st place in most VMware deployments. What I’m saying is that those of us out in the industry who are praising the virtues of NFS for VMware need to be publishing the whitepapers and architectural blueprints so that others can replicate the results and see for themselves. Instead, we have people guessing as to how VMware over NFS should be implemented and then getting it wrong (as evidenced by pkillur’s comment above), and this leaves a bad taste and sets a bad precedent. See what I mean?
I’m hoping that the new Virtualization and Grid Infrastructure BU will help lead the charge in this effort.
BTW, shoot me an e-mail when you get a chance…I have something to share with you. Thanks!
Monday, January 14, 2008 at 9:01 pm
Nick Triantos
I couldn’t agree with you more on the blue prints, architectural implementation documents and not only. My response to that, “Hold on to that thought”
Nick
PS. I shoot you an email tomorrow am. I’m babysitting my 3 kids right now and attempting to put them to bed, without much success I might add.
Monday, January 14, 2008 at 9:08 pm
slowe
Nick,
No problem, I’ll talk to you tomorrow. I just finished putting my 2 youngest kids to bed a short while ago myself.
Friday, January 25, 2008 at 2:16 pm
Wes
Now that the kids are all to bed…
Does VMware have some licensing restriction which disallows the publishing of ACTUAL numbers for various deployments? (i.e. benchmarks)
Otherwise I’d be happy to do some benchmarking on all 3 types of deployments: iSCSI, FCP, and NFS. I believe there’s a copy of Loadrunner around here somewhere…
wes
(vote ron paul on May 5)
Friday, January 25, 2008 at 11:21 pm
Eric
Wes,
The key to benchmarking ESX storage is to make sure the test reflects realistic (real world) workloads. The common mistake in this type of testing is to run a test using one guest machine, running on one ESX server. Obviously this isn’t very realistic. What I would suggest you do for your test is to run multiple guests on multiple ESX servers against a datastore. Repeat the same test for each protocol. If you happen to be testing with a decent NFS server, I think you’ll be surprised at the results.
Regards,
-Eric
Saturday, February 9, 2008 at 6:24 pm
R Robertson
As a developer of ESX, all of my VMs are mounted NFS. I’ve never noticed any performance issues related to using NFS. As you mentioned, NFS seems to be “good enough” even if it may not have the absolute best performance.