What is Enterprise-Class Storage?
To better define the term, we recall the most memorable quotes of 2006
New Year's Day is, for many IT practitioners, an arbitrary date. In many large enterprises, it doesn’t correlate with the end of a fiscal period. For example, some companies close their books as early as July or September, while others follow IRS tax calendars and set their year-end clocks to 11:59 PM on April 14. Thus, for many of us (those not in retail businesses, at least) the real value of the holidays is the "quiet time" they provide to do an accounting of our own experience of the previous 12 months.
Personally, I like to set aside a few hours during this time to make a list of the things accomplished and to close out my thinking on issues and projects that were unresolved. Doing this can clear my mind so I can set priorities and goals, and map out strategies and tactics for the coming year.
I have the time to reflect on what people have said or written to me—thoughts I found noteworthy and filed away for later review. I’ve decided to share some of those comments in this first column of 2007.
At the beginning of 2006, I was in Pittsburgh delivering a dinner talk on behalf of CA to many IT and storage professionals and chatted with a woman who managed storage for a medium-sized financial company. She was perplexed: "We have a Fibre Channel SAN with about 120 TB. I wanted to join a local SAN user group, where storage admins with local companies share their tips and techniques for managing their fabrics, but my application was rejected. The membership director said that because I had less than 350 to 500 TB, I wasn’t really doing ‘enterprise-class’ SANs."
I was surprised by her story then, and it bothers me now. Aside from the obvious egotism of the managers of this so-called user group, the assertion that storage capacity discriminates "enterprise-class" from "non-enterprise-class" infrastructure made little sense to me.
Her words have come back to me over and over again as I listened to CIOs and storage administrators in their offices or attended customer presentations at trade shows where I was supposed to be wowed by the capacities that the presenter was managing. It occurs to me that what "enterprise-class" storage actually means has never been precisely defined.
At one time "enterprise storage" meant mainframe storage; everything else, including the storage behind distributed computing environments, was not "enterprise-class." After Y2K saw the migration of almost 70 percent of critical applications off mainframes and on to distributed platforms, this distinction no longer held up to scrutiny.
What has replaced it? In my jaded view, the meaning has to do more with the price tag that vendors place on their gear than it does with any actual features, functions, or performance characteristics of the storage platform.
What Should It Mean?
Initially, the term "enterprise-class" seems to suggest that the infrastructure is somehow core to the operations of the overall business enterprise. Without this "enterprise-class" infrastructure, properly operated and managed, the company would simply cease to exist.
However, time and time again the definition of "enterprise-class" storage has fallen prey to slippery interpretations by vendors who have managed to establish a pecking order determined by vendor pricing models.
In a small business, where critical storage may include file servers, NAS boxes, internal disk, and perhaps some direct-attached or UDP-attached disk for databases and archives, the term "enterprise-class" may be justified—but vendors almost universally refrain from attaching the descriptor to products aimed at small- to medium-sized businesses. The subtle message is that while these topologies provide support for business-critical operations of their SMB users, they do so at a lower price point than what the Global 2000 might be willing to fork over for their storage repositories.
In the Global 2000 context, vendors tell us, the definition of "enterprise-class" storage may include all of the preceding topologies, but it really refers to very expensive FC fabric-attached (or FICON-attached, in mainframe shops) storage infrastructure and, increasingly, storage clusters or "grid" storage. Rather than the quaint sub-10 TB capacities of the total storage in smaller companies, big companies need big capacity infrastructure that is increasingly denoted in "petabytes," rather than "terabytes." The closer you get to the "P" level, the more "enterprise-class" your storage is—and the pricier it becomes.
Given my experience with large storage repositories, combined with some useful research from Sun Microsystems CTO Randy Chalfant, I’m not sure that capacity or price are what we ought to be using to differentiate "enterprise-class" from "non-enterprise-class" infrastructure. According to Chalfant, in another memorable quote from 2006, "Less than 41 percent of the capacity on any disk drive in any organization is being used to host what could be construed as useful, meaningful or purposeful data. The simple truth is that 59 percent of capacity of any drive is wasted, pure and simple."
If Chalfant’s numbers, derived from a longitudinal study of the results of storage assessments conducted as part of thousands of Sun’s client consulting engagements, are true, large-scale repositories may actually be more wasteful and less efficiently managed—and I should think, less "enterprise-class"—than smaller repositories in which budgetary limitations make the prospect of continuous and unlimited capacity growth something akin to a child’s Christmas toy wish list.
Obviously, moving bits between lots of spindles has inherent challenges and requires specialized skills. However, if a company has decided—as most Global 2000s have—to hitch its storage future to the wares of a particular brand name vendor, then a lot of capacity management problems go away.
Homogeneous infrastructure (all parts from one vendor or cadre of vendors) tends to be more "capacity manageable" than heterogeneous infrastructure (parts cobbled together from different vendors). The oft-repeated message from the brand-name vendors is that buying only from one vendor (the one pitching you his products) will make your life easier from the standpoint of adding capacity and moving data from frame to frame.
Given this point of view, the heterogeneous storage infrastructure common to a smaller firm may present more interesting technical challenges—from a capacity management standpoint, at least—than does the homogeneous infrastructure of a larger firm. The woman quoted at the beginning, with her 120 TB heterogeneous FC fabric, may actually need to become more knowledgeable about her "non-enterprise-class" storage and to manage her infrastructure with greater "enterprise-class" discipline than her Petabyte-sized peers. One thing is for sure: she probably wouldn’t benefit much from hanging around a club of arrogant, self-aggrandizing storage managers who have become dependent upon regular ego reinforcement from vendors who are happy to oblige, especially if it will help sell the admin more gear every year.
Just maybe, "enterprise-class" means something more than large capacity. Maybe it really has something to do with how we manage the task of managing both infrastructure and the data that traverses it.
This brings to mind another quote, this one from Yogesh Gupta, CTO and Senior VP at CA. He remarked to me once that he was sick and tired of being introduced on sales calls to a CIO of a company known as "Mr. Smith, who manages a $3 billion IT budget for company XYZ." Gupta said that he would much rather meet "Mr. Jones, who cut 15 percent out of last year’s IT budget while increasing service levels by 20 percent." That kind of CIO, to paraphrase Gupta, has "enterprise-class" written all over him.
Management, Not Size
And so it is with "enterprise-class" storage administration. I am much more impressed with a right-sized storage infrastructure that is managed according to business and application requirements than I am with an overblown SAN that has been deployed with the expectation of "fitting applications to it later."
Believe it or not, senior architects for a major automobile tire company proffered just this idea of "deploy SAN first, define application requirements later" to me earlier this year when explaining their FC SAN strategy. "Bass ackwards" as it may seem to design or deploy storage infrastructure before you understand the access, update, retention, or protection requirements of the data it will host, the approach seems to be integral to the IT management cultures of many larger enterprises I visit. It is encouraged by SAN vendors and is especially preferred in companies, such as a financial firm in the upper Midwest I visited in 2006, where the CIO has more or less abdicated his role and decided to outsource his thinking process to his vendors.
This IT maven stated explicitly, and in front of a room of people, that he would "prefer to manage a couple of vendors rather than manage any actual technology." He said that technology, especially storage, was increasingly a commodity thing, so trusting name-brand vendors to deliver suitable platforms was all he needed to do to manage his company’s enterprise infrastructure. He added that he was "close to retirement" and "didn’t want to lock horns with EMC in the boardroom"—plus, going along with the vendor afforded him "more time on the golf course."
Does "enterprise-class" actually mean "vendor-managed" as this individual suggests? Are "enterprise-class" storage products enterprise-class because of who sells them or because of what they cost, or does "enterprise-class" have something to do with value-add software added to an otherwise commoditized white box array?
Increasingly, vendors assert that the latter is the case. A spokesperson for one up-and-coming storage vendor, 3PAR, told me recently that "performance no longer matters in the storage sale." His customers don’t care about product performance, as long as it performs at minimally acceptable service levels. He said that companies are buying products that have a lot of nested functionality to automate the archiving (a.k.a., hierarchical storage management) and data protection processes delivered in a single, integrated cabinet.
What makes his solution "enterprise-class" is its "value-add" functionality, plus a simple capacity scaling method (using only his company’s trays of commodity drives), served up as a one-throat-to-choke maintenance contract. While this sounds very much like a pitch from EMC, IBM, or HDS, the fellow’s words cannot be discounted out of hand.
Capacity, according to many vendors, doesn’t matter one iota in the definition of "enterprise-class," provided the "solution" will "scale seamlessly." Automating routine administrative functions on the platform is also important and facilitates the management of increasing capacity without a commensurate increase in administrative personnel requirements.
Of course, value-add functions contribute to a pricing model that discriminates one class of product from another. The more automation you add, the more expensive the platform (from an acquisition standpoint, at least; we can argue later about soft cost savings in the areas of labor, electricity, etc.) and the more "enterprise-class" it becomes.
Is this generally accepted industry wisdom—more automation equals greater enterprise suitability—really a definition of what constitutes "enterprise-class" storage? I don’t think so, and neither should you.
What makes storage "enterprise-class" is what you do with what you’ve got.
Are you protecting the data in a sensible way? Are there routines in place for backups or mirrors? If you need to retain data, is this indexing and archiving accomplished and accounted for in a policy-driven way? Are you continuously checking to ensure that your storage investment is right-sized to business process and application requirements? Are you grooming your disk on an on-going basis to separate the grain from the chaff?
The answers to these questions illuminate what "enterprise-class storage" really is—regardless of the aggregated capacities of the spindles you have deployed, the brand names on the boxes, or the feature/function software bundled in the purchase.
In the final analysis, enterprise-class storage is a verb. It is not a noun.
Your opinions, as always, are welcome: firstname.lastname@example.org. Happy 2007.