There has always been a lot of debate in the VMware community about which IP storage protocol performs best and to be honest I’ve never had the time to do any real comparisons on the EMC Celerra, but recently I stumbled across a great post by Jason Boche comparing the performance of NFS and iSCSI storage using the Celerra NS120, you can read this here.
What you’ll find reading the later part of Jason’s post is once a few tweaks were made for NFS the results were actually very similar. So if there is not a clear winner on the day, how do we best decide which is the best storage protocol for your VMware environment ?
The “Bigger Picture”
I can tell you in the past I have always deployed Celerra’s using iSCSI and I like to think this choice was made with the “Bigger Picture” in mind. If we go back in time and look at VMware support matrix’s you’ll notice that a lot of the add ons such as VMware Consolidated Backup, Storage vMotion, Site Recovery Manager all supported iSCSI well before NFS was officially supported.
It was these considerations early on that lead me down the iSCSI path, and then of course later on iSCSI was something I become comfortable with so naturally it become my protocol of choice.
Another consideration with the Celerra was integration with EMC’s Replication Manager which could be used to provide application consistent snapshots of Exchange, SQL, Oracle and VMFS datastores when iSCSI was used.
That was before, how about now ?
So a couple of years down the track and things have changed considerably, VMware Consolidated Backup, Storage vMotion, Site Recovery Manager 4.0 and Replication Manager 5.2.2 now all support VMware NFS datastores.
Ready to change to NFS yet ?
Even with all these changes, I still was not ready to move away from iSCSI to NFS because vSphere 4 brought major improvements to the VMware software iSCSI initiator which allows multiple VMKernel ports to be bound to the iSCSI initiator to give the ESX host multiple paths so the storage array.
Shame on me
So earlier on in the post I talked about the “Bigger Picture“, Is the improvement toVMware’s software iSCSI initiator part of the bigger picture ?
No, this is a small technical nice to have feature, but really I needed to take a step back and think about what NFS means to Celerra and what makes NFS appeal more so than iSCSI. (Keep reading to find out)
Where NFS trumps iSCSI on the Celerra
Replication Manager— When creating an iSCSI lun you first create a file system, think of the file system as a container for the iSCSI lun.
Without the need for snapshots, this file system only needs to be fractionally bigger then the iSCSI lun to accommodate for meta data, but as soon as you need to start performing snapshots of iSCSI luns, the requirements for additional file system overhead change completely.
Long story short with a fully provisioned iSCSI lun, the minimum file system space required to perform a snapshot of the lun (and not taking into account changed data) is 2 x the published lun size, there are of course ways to reduce this required overhead and if you want to read more about this you can read one of my older posts about this here.
Replication Manager 5.2.2 as mentioned earlier, now supports snapshots of NFS datastores, the good news here is the Celerra uses a totally different method for the snapshots of NFS file systems (using a dedicated savevol rather than its own file system) and allowing 20% overhead for snapshots is a realistic figure.
Site Recovery Manager— VMware Site Recovery Manager using Celerra Replicator also uses the Celerra snapshot functionality to replicate Source iSCSI luns to a remote Celerra. The same overhead requirements as noted in the Replication Manager section are applicable here also.
Site Recovery Manager 4 of course now supports NFS, existing customers feeling the pain from the overhead needed to support iSCSI snapshots can at least now migrate everything to NFS datastores and claim back a tone of that valuable capacity.
EMC Celerra I beleive was the first storage platform to provide automated fail back, this is done using a vCenter plugin, the new version now supports NFS.
Celerra NFS vSphere Plugin— EMC recently released a new Celerra plugin which integrates with the Celerra, you really have to check out this YouTube video to see how cool some of the integrated features are when using NFS.
Deduplication— EMC Celerra has supported deduplication for some time now, almost every release of DART recently has brought with it considerable improvements. The one that’s caught my eye in the latest release 5.6.48 is now optimized to work with VMDK files in an NFS data store, reducing space consumption by up to 50 percent. The overhead of accessing the compressed VMDK is ess than 10 percent. When using the Celerra NFS vSphere plugin an administrator can select which virtual machines in the data store to compress. If the additional overhead is too much then the administrator can simply uncompress the virtual machine on the fly.
The point of this post is not to advocate NFS over iSCSI, my intention here is really just to show how important it is to take that step back and look at the overall solution before you rush ahead choose a protocol which may not end up being the best choice for your environment.
As a consultant who implements systems, reviewing the two different protocols was a good reminder to myself not to get too stuck in my ways, things change !