VMware Makes NFS Mainstream
By Tony Asaro on Jul 16, 2010 | In Virtualization, Storage, Storage Management | 6 feedbacks »
We did research on IT professionals and their major priorities over the last 18 months and all of them emphatically stated that implementing VMware was one of their top initiatives significantly driving down costs and improving operations. These IT professionals also said, however, the biggest challenge they faced with their VMware environments is network storage. Dedicated storage managers are buried under a mountain of projects and day-to-day operations that consume so much of their time that taking on new tasks is nearly impossible.
Another important dynamic we’ve observed is that VMware administrators are finding themselves becoming quasi-storage administrators, which is somewhat akin to asking a brain surgeon to operate on the heart. This is not due simply to the inherent complexity of most networked storage systems; it is an issue of specialization as well.
Ironically NFS, the protocol of geeks, is a means to simplify VMware and networked storage. It turns out there are a number of customers using NFS with VMware and as a result they have eliminated much of the day-to-day complexity of managing network storage with VMware. The following are the advantages of using NFS with VMware versus FC and iSCSI SAN approaches:
• It’s very simple to add NFS datastores. With NFS there is no LUN management. This simplicity addresses configuration issues that might come back to bite you in a SAN environment. The storage provisioning process for SANs often requires a dozen or more time-consuming steps. Missing any step may result in disaster and you might not know it until it happens.
• You don't have to deal with all of the complexity of Fibre Channel, WWNs, zones, ISLs, etc.
• You can easily increase or more importantly decrease the size of NFS datastores online for capacity reclamation.
• Since you are using NFS you don't have to deal with VMFS or RDMs. This means you can have volumes bigger than 2 TB. NetApp supports a 16 TB file system; Isilon and BlueArc can support 100s of TBs in a datastore (and theoretically more).
• There is no single disk I/O queue with NFS, which means that performance is dependent on network bandwidth and the storage system itself. It also means that NFS performance can keep up with even FC in VMware environments.
• You can backup whole VMs, or files within VMs.
• Restoration of VMs is flexible including individual VMs, multiple VMs or files within VMs.
• The cost of FC is higher including the equipment and support. This is an argument for both NFS and iSCSI over FC.
• One customer pointed out that the tools to troubleshoot IP networks are much better than FC, another advantage for both NFS and iSCSI.
• SAN expertise is more specialized than IP – it is harder to find the experts and to retain them (and they usually paid more).
The problem is there aren’t many vendors that offer storage systems that support NFS. There are literally dozens of SAN-based storage systems but only a handful of NFS or NAS storage systems in the market. Customers want choices and the fewer there are the less likely they will go down a particular path. And there are good reasons since competition fosters innovation, cost effectiveness and better service.
NFS for VMware has the potential of changing the storage landscape, but sadly, it probably won’t. The reasons are: it is a major challenge to educate the market; there is an enormous amount of incumbent SAN storage systems; and there is no one taking up the mantle to fight this good fight. Even NetApp has lost its fervor for NFS and has instead taken a “we provide whatever protocol you want” attitude, responding to the market versus driving it. That isn’t to say that NetApp isn’t promoting the use of NFS with VMware but it is not a core strategy and is only one of many things they talk about. BlueArc and Isilon are now focusing on NFS for VMware but they don’t have the same level of awareness or resources as the big storage vendors. However, if NFS for VMware is going to take off, these two vendors are in a position to sow some seeds and reap the rewards.
NFS is a more highly virtualized protocol than FC and even iSCSI. As a result it works much more easily and efficiently with VMware. If NFS is to become the dominant VMware protocol it is probably not the storage specialists that will make this happen but rather the VMware administrators who want highly virtualized networked storage without having to be a storage expert. The implications of this are significant as VMware continues to proliferate and those that manage these environments influence and decide what they need to be successful.
6 comments
One more benefit of NFS that we learned the hard way when we lost 50+ production servers in a single instant.
No SCSI reservation conflicts!
Because of a bug in a well known vendor's storage monitoring software, we encountered a SCSI reservation conflict on our FC SAN which caused all of our vSphere Enterprise Plus hosts to lose connectivity to the SAN.
With help from VMware support we were able to recover the primary ~2TB VMFS volume that contained all of the VM images within about 2 hours.
As for the ~4TB of RDM's that contained all of our unstructured file shares, we had to recover from backup, a 48 hour outage in all. Not a pleasant experience!
Needless to say, we are planning to move to NFS. As you mentioned, many of the storage vendors aren't even aware of the issues with VMware and SAN storage architectures. They have that “deer in the headlights” look when we ask for an NFS storage solution for VMware.
It would be interesting to see how you progress on the NFS front. Obviously NetApp provides a solution and I know that BlueArc and Isilon are focusing heavily on VMware as well. Which vendors are you looking at?
Tony
The leading candidates at this point are Compellent and NetApp. We are beating the bushes to find other players. We'll take a look at BlueArc and Isilon.
Thanks,
Gregg
We have been debating migrating the entire enterprise onto VMware but are undecided at this point. We are probably the right size to pull it off with less than 400 employees, moderate email traffic and no heavy OLTP. However, we had performance issues trying to run Citrix on VMware 3.5 supporting ~120 users a few years ago so there is some hesitation within the organization toward moving resource intensive apps/servers back into that environment.
Part of the issue back then (2006) was our lack of understanding of the horizontal scalability paradigm of VMware. We were only running a pair of VM's on each of a pair of ESX hosts. This config resulted in 25-30 concurrent users per VM during peak periods. That wasn't a problem except when a rogue process hung one of the VM's or consumed massive amounts of memory, which happened too frequently on that version of Citrix with the application mix that we were running. Knowing what we know now, we should have been running 4 or even 8 VM's on each server and it probably would have been a fantastic combination.
Leave a comment
| « Dell and 3PAR = No Brainer | The Emperor Has Too Many Files » |

