Who would have thought such a simple question – really more of “seeking to understand” as my VP of Sales Mark Glasgow calls it – would kick off such a slew of e-mails, comments and tweets. The other day I asked the following question on Twitter:
@StorageTexan: question 4 all NFS/NTAP/Celerra Luving ppl. wht is ur opinion on y NFS is superior ovr block device access in #VSphere?
I got a TON of feedback, so much so that I decided to ask the question on my blog as a means to collect the responses and to allow others more than 140 characters to respond Out of that came some really interesting responses, including George Crump sharing his views in Information Week.
Let’s take a step back. The majority of my storage life has been in the block-based access protocol. For the majority of my career it was all about Fibre Channel connectivity. Have you heard the saying “If all you sell are hammers, then everything in the world is a nail”? That’s sort of where I was at 5 or so years ago. The FC was my hammer, and all connectivity issues could be resolved with FC!! Then a few years ago iSCSI started to inch its way into the discussion. Xiotech adopted it as another option for connectivity and I got to add another tool into my bag
Then I asked the above question as a means of self reflection. Why would someone choose to bypass block-based connectivity in lieu of file-based? Just didn’t seem logical to me. I even happened to have that particular tool in my bag, so it’s wasn’t as if I was trying to compete against it, just wanted to see what the big deal was. Today, almost all storage vendors (Xiotech included) offer NFS connectivity. Some utilize gateways like the EMC Celerra, NetApp V-Series as well as Xiotech. Others use native support like PillarData and NetApp FAS product line.
At first blush I thought, it has to be because IP is viewed as being less expensive, less complicated then native Fibre. But I think at this point, the price argument against FC should be put to bed, thanks in large part for Cisco/Brocade/Qlogic driving down the costs. Complexity is also something I think should be/could be put to bed. Just sit in front of a Cisco Ethernet and a Cisco MDS switch, the IOS is the same; the perceived complexity around FC is really no longer an issue. Now, for the smallest of the SMB, maybe costs and perceived complexity is enough to choose NFS. I can see that.
Maybe it’s because these gateway devices offer something their block based architecture can’t support. That starts to make sense. Maybe it’s some sort of feature that drives this decision. In some cases, maybe its thin provisioning, better integrated snapshot, single instance storage/Data DeDupe and even advanced async replication. Most storage arrays on the surface can do these, but with a gateway device maybe they can do this better, cheaper, faster, etc.? For the SME/SMB I can see this as a reason.
Then again, according to some of the people that responded to my blog and twitter, maybe it’s for performance reasons. Some sort of ability to cache the front-end writes make the applications/hypervisors/OSes just run quicker. Others suggested that gateway devices made it just “stupid simple” to add more VM’s because you could essentially treat an NFS mount as a file/folder, and you can just keep dropping VMDK’s (files essentially) into these folders for easier management. That makes sense as well. I can see that, I mean if you look at a NAS device it’s essentially a server that runs an OS with a file system that connects to DAS/JBOD/SBOD/StorageArray on the backend right? It could be viewed as a caching engine.
Then it dawned on me, it’s not really about one being better than the other, it’s more about choices. That’s what “Unified Storage” is all about, the ability to add more tools to your bag to help solve your needs. If you look inside a datacenter today, most companies have internally tiered their applications/servers to some extent. Not everything is run on the same hardware, software etc. You pick the right solution, for the right application. Unified Storage is the ability to choose the right storage connectivity for the various different applications/hypervisors and operating systems. The line gets really blurred as gateway devices get more advanced and better integrated.
Either way, everyone seems to be moving more and more to the Unified Storage device. It should be interesting to see what sort of things come out of Storage Networking World in a few weeks !!