Tiers, Tiers Everywhere

I've been noticing increased activity from vendors large and small to offer newer tiers of storage in our crowded, noisy marketplace.

That's a good thing -- the more choices we all have, the better.

But, at the same time, I think it's also important to have a bigger context about what's going on here, and what it might mean for consumers of these technologies.

Basic Concepts

We use the concept of "tier" in storage as a shorthand for a distinct set of tradeoffs: cost vs. performance, cost vs. availability, cost vs. recoverability, and so on.

Different combinations of technologies create "tiers" of storage.  Given the wide range of storage requirements in most organizations, there is no best "tier" -- the trick is to use as few tiers as possible that meet the organizational requirements for service levels, while keeping cost and complexity under control.

EMC started thinking in this way back in 2003 -- it was a predecessor concept to our thinking around ILM.

New Tiers Abound!

I don't know if you've noticed, but there are many newer options abound for new combinations of tradeoffs in storage.

Sure, there are disk drives of different geometries, form factors and capacities.  Always new choices there.

There are different ways to use these disk drives; different levels of RAID, dedupe approaches, thin provisioning, scatter data geographically, and so on.

And there are interesting alternatives to physical disk drives: enterprise flash drives, RAM configured to look like a disk cache in various ways, and we'll see a lot more of this soon, I'm sure.

And, to be fair, our old tired friend tape still plays a role in the tiering scheme.  As do paper printouts, microfiche, optical, and other things that we usually don't think of as "storage".

Understanding The Tradeoffs

Every storage technology is an exercise in tradeoffs.

This should be no surprise to anyone, but I've seen that sometime people's enthusiasm get in the way of this.  Sometimes this is driven by normal vendor enthusiasm for their particular way of arriving at tradeoffs; other times it's technology watchers who get smitten by a new toy.

From my perspective, EMC is pretty conservative (as far as vendors go) in tempering our enthusiasm for any one particular tier or technology over another.  We take great pains to position different tiering options (e.g. enterprise flash, terabyte drives, CAS, etc.) in terms of tradeoffs and use cases.

I guess we have to adopt this perspective -- we've got so many different (and potentially attractive) ways of storing / replicating / recovering information.

What Should Customers Do?

I've discussed this before, but it's worth bringing up again in the context of this discussion: establish your storage service level catalog. 

The concept is relative straightforward -- create "classes of storage service" (e.g. tiers) in terms of performance, availability, recoverability, etc. -- and expose the true costs of providing those storage services.

The fewer you can get away with, I'd offer.  I've seen a range of 3 tiers at the very low end, and as many as 20 for very large, complex environments.

So, why is doing this important?

Several reasons, but in the context of this discussion, it makes evaluating new technologies (e.g. tradeoffs) much, much easier.

When something new comes along, you can look at your service level catalog and say "does there now exist a better way to provide this new service level?".  Maybe there is, and maybe there isn't, but it certainly helps to understand if there's a new answer if you've got the questions identified already.

The Complexity Factor

The more tiers, the more complexity can be an issue.  There are several strategies I've seen to keeping the complexity manageable.

First is the growingly popular concept of multi-tiering in the same frame.  Many products (including several from EMC) can offer up a wide range of service levels in the same frame simultaneously.  Features that identify potential optimizations (moving information up or down a tier) are useful, as well as the ability to do this without significant application impact. 

It's a relatively new idea, but growing in popularity.

Larger organizations who aren't comfortable with this approach may use the same hardware platform, but have different configurations optimized for performance, availability, cost and so on.  They all manage the same way, they all have common capabilities.  We see this quite frequently at EMC.

Other organizations prefer to implement their service catalogs with different products from primarily one vendor.  This gives them some common capabilities across the different platforms (e.g. certain aspects of management, replication, etc.) but -- more importantly -- one vendor to deal with.  My impression is that -- far and away -- people who pursue this approach do so with EMC.

On The Down Trend

Over the last few years, the classical Gartner-esque "you should be a multi-vendor shop for cost negotiation reasons" has been on the decline, as least, as far as I can tell.

I think the reason is simple: any (potential) savings you might get from this approach are far outweighed by the costs associated with additional complexity.  Every additional vendor brings a very real cost: someone else to deal with.  Same sort of reason we all tend to use the a single email platform, ERP package, etc.  And to the extent that storage is more about software these days than hardware, it makes a certain degree of sense.

Now, I'm not saying that all of us vendors are to be completely trusted in all circumstances, but, please keep in mind, we all operate in an extremely competitive marketplace, which tends to keep all of us on our toes, even if we're the primary vendor.

So, What Do I Think?

I think this is a logical extension of "features vs. products" argument I've shared before. 

In a world where vendors are trying to make it on a single capability (insert long list here, including dedupe, thin provisioning, drive spin-down, cached NAS accelerators, sealed disk canisters that - hopefully - never need service, and even file acceleration for AutoCAD files), we should keep something in mind.

It's all part of a bigger picture.

Comments

[Peter Kraatz]

I agree with the central theme of "less is better" and the approach we take in the EMC consulting practices expands upon that theme somewhat by saying "get granular, then reduce the grains..."

We do this by dividing the "storage" into its purposeful buckets by Service Area- Primary Storage, Operational Recovery, Disaster Recovery, and Archive Storage. Each of these Service Areas is tiered with a catalog and reference architecture to support it, so while Primary Storage may have 3 tiers, Archive could have 7, D.R. might have 1, and Operational Recovery (backup) 4- all based on business requirements and nothing more.

It is the business requirements that should drive the tiering design and the storage team should be building the smallest number of solutions to meet ALL the requirements. As you point out, the same technology solution can be applied to multiple problems, minimizing complexity.

I am of course oversimplifying for space here, but the "new toy" illustrated in your link above highlights a common error of thought in the marketplace (and one you've alluded to): that all storage solutions have something to do with disk, and primary disk at that. Hogwash. If it were true, we'd have no filing cabinets left.

Only by breaking down the problem according to business requirements and service functions can you really solve the problem in a lasting way.