I find it interesting and telling when a vendor tries to put another vendor’s product into a niche; sometimes it’s justified but oft-times it isn’t. It’s even more interesting when a vendor tries to put their own product into a niche to defend the market position of one of their other products.
Currently watching some of the positioning of Isilon both from NetApp and EMC themselves is amusing; Isilon is currently being positioned as a ‘Big Data’ solution by EMC and some confused witterings around it’s applicability to general purpose storage from NetApp are helping muddy the water.
And I can see similar issues around other Scale-Out products; where positioning is at times very defensive to prevent cannibalisation of existing products; for example, IBM currently find hard to position SONAS versus nSeries. Their current positioning of nSeries for less than 100TB and SONAS for larger deployments is simply marketing masquerading as technical strategy.
Quite simply, if you are looking to put in place or refresh your file-serving environment in the next couple of years; you are doing yourself and your employer a massive dis-service if you do not look at a Scale-Out solution. At present, they do not have some of the features of the traditional dual-head design; deduplication and compression come to mind but those features will come but architecturally they are designed to scale and if we know anything today, scale is going to be important to everyone.
They do this elegantly and seamlessly; adding additional capacity is simple and transparent; migration and maintenance is simple; over the years, this will save you probably more money than de-duplication will today.
The next thing will be to see if someone can bring a equal level of simplicity to block….
[Edited to show that I do understand IBM’s SONAS positioning but think it’s less than credible]
As is normal nothing is perfect nor the best at everything and this includes the Isilon product but where such products do fit they can do an excellent job and as you say people should take a serious look at them though as usual you need to be very clear on what you are looking to achieve and not lose focus on that.
I think that there is more chance that the Scale Out products will do what you want to do tomorrow as opposed to the non-scale out which will potentially force you into a silo-ed and overly complex infrastructure which will become more and more painful to manage and mitigate against shortcomings.
Adding scale-out to silos is incredibly complex and a fools own errand for many vendors.
Disclosure – EMCer here.
‘Bod – hope you’re doing well – found this blog post fascinating. First of all – while EMC positions Isilon as “Big Data” (most beneficial to customers when their scale or use case demands scale-out), there’s NO DOUBT that scale out architectural models have inherent advantages, and are widely applicable in broad use case scenarioes.
In fact – I personally would go as far as to agree 100% with your assertion – “…if you are looking to put in place or refresh your file-serving environment in the next couple of years; you are doing yourself and your employer a massive dis-service if you do not look at a Scale-Out solution…”
I do want to make two notes.
1) I’ll pile on the observation that certain “classic” enterprise NAS features (file-level snapshots, large scale filesystem writeable snapshots, compression, dedupe, etc) are currently not there in Isilon/SONAS and their ilk. These are NOT instrinsic, and I completely agree that adding those features is likely to prove easier than to “bolt on” than it is to “bolt on” the core architectural model. Expect a lot on this front.
2) With scale-out NAS models, individual IO latency is… TODAY… generally material higher for transactional NAS workloads. This is based on distributed metadata and data models, and the inherent latency that introduces. Just speaking for EMC, the VNX file stack has much lower small-block IO write latency (for things like VMware on NFS, or Oracle on NFS) than Isilon. Scale out block models (3PAR, VMAX, HDS) use very specific hardware models (generally shared memory) to achieve low IO latency while keeping scale out working. This is the “harder” area of R&D when it comes to scale-out NAS (as the inherent architectural benefits of scale-out instrinsically make latency trickier – the classic “your strengths are simultaneously your weaknesses”). This is an area of insane work. If it can be cracked, the possibility of scale-out NAS models to become extremely dominant will be much more pervasive.
Stay tuned…
Thanks again!
Chad,
thanks for popping by, as always my opinions/thoughts are a moving target. And I tend to agree with you especially with the regards to latency in the current scale-out models and as you say, this will improve over time. I think if I was looking at a NAS requirement today, I might be looking at a tiered NAS environment but always keeping the end goal in mind which must be some kind of unified NAS tier satisfying as many of my requirements as possible.
So today, I might well be tempted to deploy NetApp, VNX in a tactical manner but manage my investment very carefully there and deploy scale-out solutions for certainly the bulk of my unstructured data. I would be asking questions of my vendors on how they can support me in ensuring that getting to my end-game is as painless as possible.