Bob Madaio

You Do Know What "All" Means, Right?

Blog Post created by Bob Madaio on May 27, 2014

When I started this series on Flash storage with the initial post on Flash Storage: Redefined that talked about our flash shipment numbers, I wanted to cover a few different topics that brought the focus on why I thought our flash shipments were doing better than many may have expected.


In the post The Hard work Of Flash Storage, I already talked about how Hitachi made special hardware and software engineering efforts to ensure that our flash storage was as easy for our customers to adopt as it was high performance, and enterprise-ready. Oh, and I patted ourselves on the back from some great public benchmark results. (And for being one of only a few vendors to reliably complete them, helping customers get a better picture of what a system's performance is really like.)

But there’s another topic that hangs around like a fog, which I thought would be good to begin addressing. The idea of the mysterious “All Flash Array.” Now, there's nothing mysterious about a storage system that is all flash. But aside from literal translations of "all flash" of a system that uses only NAND memory in one form or another for its persistent storage, this term has taken on some near mythical meaning.

To many, this term has come to include in its meaning “a system that was designed to never support disk drives.” Of course, I would offer that this should instead then be called something different, maybe an “Only Flash Array” (OFA). Aside from this only-flash connotation that many have added to the idea of “all” flash, there are smatterings of other must-haves a storage system must include, a particular feature, a particular form factor, and the odd belief that only something designed recently can process IO to flash correctly.

This topic has come to the forefront once again as everyone’s favorite storage guilty pleasure trade publication, The Register, has published two articles on the influential analyst firm Gartner’s upcoming publication of a flash storage “Magic Quadrant” a mainstay of the firm's comparative research. Article one sets the stage of what The Register expects from Gartner, whereas article two vents one vendor’s (Dell) frustration about the selection criteria.

Astute readers of those articles will note that Hitachi is also not included in that list either.  I don't intend this blog as some sort of response to Gartner, but rather a quick view of our perspective on all flash storage.

At Hitachi, we have engineered our systems to perform splendidly when deployed as all flash arrays. We don’t offer a system that is an Only Flash Array. For any comparison (Gartner or otherwise) that spec’s out systems which don't conform as not “non-upgradeable” (downgrade-able?) to include disk-drives, we would be disqualified as per semantics despite having all flash configurations. Could we offer an OFA-only model someday? Sure, but let's look about why we aren't clamoring for this.

We don’t get OFA-only concerns from customers. Customers ask for a solution, and we offer one that responds to their needs.

Many want tiered systems, with our Hitachi Accelerated Flash storage (or SSDs) being a portion of the system, and HDD capacity being there for tiering and data that is lesser used or has a lower performance requirement.  To these customers, it just makes economic sense to have both. And when Hitachi Dynamic Tiering is used, we’ve seen large firms run sub-millisecond average response on large tiered databases.  Customers like that see no reason to worry about all flash at this current moment.

On the other hand, we have a growing number of customers whose applications must never miss the low latency of flash and hit relatively slow disk drives.  Of course, despite the economic advantages, tiering cannot guarantee that an I/O won’t be processed from the HDDs.  For these customers, we propose all flash solutions.

In an all flash discussion, customers have not seen our ability to offer the same system to include disk drives as a negative. For many all flash customers, it is irrelevant since the solution we’d propose to their current problem would not offer HDD. (E.g. the customer that says “I’m all flash, and that’s that.”)

What's relevant to that customer is the performance, application and hypervisor integration, functionality, manageability, reliability and economics of the overall solution, relative to what other vendors are proposing. And when customers focus on these metrics, we do quite well competing against the OFAs of the world.

There are, as it turns out, real world benefits of having a system that is as fast (or faster!) as the Only Flash Arrays, yet capable of including disk drives. The first benefit, is that of flexibility and avoiding “buyer’s remorse.”  Some customers appreciate the idea that if we install an all-flash system, and the environment/needs change, we can convert that system to tiered with HDD. (E.g. the customer that says “I bought this as all flash since I knew what 20% of my data needed to be on flash. But you know what, that 20% isn’t static. Instead of me worrying I’m getting the right data on flash, let’s tier this system and let the storage do that for me.”) This is one reason that customers have occasionally virtualized OFAs behind our storage controllers to ease management and allow the system to treat it like a tier.

The other type of customer that really seems to appreciate systems that can be optimized for all flash, or deployed as tiered with HDD (or, yes, still some that are all HDD), are those that know right from the start that they will run applications that require all flash, many that will require tiered systems, and maybe a few that won’t need any flash (like a pure “tier-3 performance” tier in a virtualized storage environment.)  This type of customer appreciates that across our management software, replication software, application and hypervisor integration, encryption, etc., etc, there is complete commonality.

Because it turns out, while many customers need all flash arrays, not that many environments are only-flash yet.  We can help with either, and I’ll be back on the blog to talk even more specifically about how we compete in an only-flash configuration for those customers who are “all in.”