Spinning The Hype for All-Flash Arrays
Our industry will latch on to something and use marketing tactics to try and establish a trend, gain mind share and ultimately sell products that the consumer does not always need as badly as advertised. It's not a bad thing, it's how the market works. After all, companies are in the business to make a profit. That's where analysts like me come in, to dissect some of this hype and present you with the facts.
So, the only word I could find to properly describe the craziness that is surrounding all-flash arrays and the amount of hype around it is "epidemic."
I don't mean to sound as if I'm anti-all-flash arrays, not at all. But I must insist that all-flash arrays are not the be all and end all that will fix all of our problems and it most certainly will not kill spinning disks. Our industry will gravitate more towards hybrid solutions while rethinking existing storage architectures and approaches. So to be clear, I do believe that for certain point solutions all-flash arrays are well suited, but for the majority of workloads, I'm not so sure.
Recently I visited with a large customer who also follows this blog and learned a lot about how their business is evolving and how IT is closely and intimately aligning with business goals and the business in general. It reinforces what I've been saying and repeating to anyone who will listen: "There are no IT projects, only business projects and initiatives." Back to that conversation, my customer also told me that they now live and die by asking, "What for?" Everything they do, every product they purchase, every project they initiate, they ask, "What for?" The question reminds them of why are they making the decision they are making and how that decision will translate back into the business.
Now, back to the flash problem: The first thing that I want to bring up is warranty and how flash arrays stack up against traditional and hybrid solutions from a warranty perspective. If you have been following my blog you know that I place a high importance on the role of procurement moving forward. Not that their role was not as important up until now, but just because I believe in the age of cloud they will be scrutinizing the purchases, squeezing them and ensuring the IT department has thought of all the aspects that are not always technology-related. So, for starters, if you are considering all-flash for any reason, I would be wondering about the warranty.
Most procurement departments are expecting three to five years of service out of their IT-related equipment. Vendors that come in with products carrying 1- or 2-year warranties will find the bottleneck with the procurement department, not the IT department and that is a business factor that you should never discount.
My favorite line about all-flash arrays is that they are faster and perform better. While they are undoubtedly fast, to come out with a blanket statement that they are faster is misleading. Sure, flash arrays significantly reduce latency, which is crucial for many applications. But have you considered the bandwidth requirements for certain workloads? Flash arrays will give you hundreds of thousands of IOPS but have you considered the profile of your workloads? Have you taken the time to understand the ratio between sequential reads and writes and random reads and writes? You might be surprised to find that flash arrays will perform exceptionally well for reads in general but might be challenged as far as random writes and even sequential writes in some instances. VDI anyone? Some might be surprised to find that for VDI I favor hybrid as opposed to all-flash for performance reasons, among other things.
The second thing that even had me going for a while was the idea that flash requires less power and cooling. It's true if you consider raw flash modules and it might be the correct choice for small environments or limited deployments. However, as this industry is touting the death of spinning disk and the era of all flash, I am making the assumption that all flash companies want to replace every piece of enterprise storage out there. When you consider this endeavor, you will find that flash arrays at scale will require additional components like CPUs and cache that will draw more power and require more cooling than hybrid systems. Again, this point is interesting -- for a limited deployment or a specific workload, it's true; at scale it's not true at all.
I could go on about some other challenges with all flash. My point is that in my experience no technology has ever been able to kill off another technology. You can provide value, you can improve but you cannot complete abolish the other technology. So, another "what for" that organizations should be asking is, "What do you really need all that performance for?"
Knowing that the average workloads in enterprises run anywhere between 50,000 and 100,000 IIOPS (and I am simplifying it here for the sake of conversation), do you need an array that gives you 1 million IOPS? If you ask me I would say yes, you can never have enough IOPS (j/k) but procurement might disagree and you always have to have the business needs and requirements in front of you as you tackle this and not just address it from a techie's point of view.
I realize that the storage world will converge on me with criticism or praise due to this blog and that's OK; I am really hoping for good conversation in the comments section.
Posted by Elias Khnaser on 11/20/2013 at 1:18 PM