In-Depth
Speaking Plainly about Purpose-Built Storage
Resellers should spend more time helping customers to spec out what they want their infrastructure to look like over the next few years.
This past week, a reseller friend was complaining that his customers are deferring purchases indefinitely. Gone are the days, he said, when folks would buy 50 TB of storage in advance of their needs. "We are getting orders for drive shelves, but we are being asked to leave them unpopulated with drives so that we have the chassis space if we need to add disks later, but we can buy them down the road."
Much as I tried to feel my friend's pain, I was actually delighted. I don't want his business to flounder, but I had to say that hard times foster innovation. The customer needs to watch his wallet right now, and the reseller is going to have to adapt to the new realities. Instead of driving the customer to "horizontal" storage infrastructure -- generic arrays designed to solve everyone's problems with embedded software features, and therefore solving no one consumer's problems very well -- perhaps he should think "vertically" -- tailoring solutions to the customer's specific here-and-now requirements.
My advice to consumers has always been to purpose-build their storage infrastructure. That, in plain spoken parlance, means: understand your business process and application, the data it generates and uses, then design a platform to provision the data with exactly the services and resources that it needs. Nobody seems to listen to this, of course, until times get hard and budgets get lean. Then, they do.
The challenge to my purpose-built storage philosophy is two-fold. First, many vendors have spent a great deal of money to make their platforms sound purpose-built. A few weeks ago, I spoke with Lee Johns, director of marketing, entry storage and storage blades, with HP's StorageWorks group, ostensibly to get some information relevant to this column, and found that we were speaking different languages.
Johns offered that his products for small and mid-size businesses (SMBs) were indeed what I would call "one size fits most." He referred to them as "a complete software stack at a lower cost, which defines value in the SMB market." He was mainly referring to the iSCSI storage from LeftHand Networks, acquired by HP late last year.
Using my terminology for purpose-built storage, Johns posited his definition of "enterprise" storage solutions. Different requirements present themselves when building an "enterprise" storage platform, he said. Storage needs to be separately tuned to the needs of applications, he noted and I agreed. Storage associated with different applications may need to scale independently of other storage associated with other apps, he offered and I agreed. Management was a "must-have" across all of these specially tuned and independently scaled storage platforms, he concluded, and again I agreed.
I was about to warm up to HP in a way that I hadn't for years when he began to describe the details of his "purpose-built" storage vision. Basically, he was suggesting that purpose-built means buying a ton of storage that can be allocated and de-allocated by an intelligent blade server head to various applications: "It's great for virtual server environments and easier to manage than multiple disparate storage arrays."
He noted that his customers were going crazy because "the cost for a competitor's storage array was greater than a server in a virtual server environment." In his view, the right architecture is a high-density JBOD built onto a server with 30 to 100 SAS/SATA-connected drives and parsed out using generous doses of PolyServe, HP's clustered file system.
He said that HP had worked with Oracle to create a "database machine" -- a specialty server-storage rig using Infiniband interconnects to provide the perfect fit for an Oracle environment. That was purpose-built any way you slice it.
I offered that a custom-engineered platform for Oracle sounded purpose-built, but it wasn't exactly my definition. To my way of thinking, purpose-built storage starts with building block resources and services that can be inventoried and monitored and managed in a common way on an ongoing basis. That gives you a menu of services that can be leveraged to create "routes" for data to take based on higher-level, business-facing resource provisioning and protection policies. Based on business rules, a policy engine defines a route through the infrastructure that will expose the data output from the application to the right set of functions -- encryption, replication, maybe de-duplication, archive, etc. -- as dictated by business and application requirements.
Moreover, purpose-building means placing functionality where it best serves the data. Some storage-related functionality needs to be close to the storage media itself, but many of the functions being embedded on arrays don't need to be placed there. They could be more economically delivered as part of a server utility set, or on a networked router engine or appliance. In fact, the more embedded value-add features we could remove from the array, the better, I suggested, since all of the value-add software that most vendors have been adding to their storage arrays merely increases the cost of the array despite the fact that the price of disks in the array keeps falling year after year.
He replied with what I normally receive from storage-systems builders: "Customers don't want to understand the complexities of storage." He said that if the customer had some custom parameters that he wanted accommodated on an HP StorageWorks platform, he could use HP's Active Answers, a Web-based configurator, to specify them. An engineering team would work on the specifications and get back to him with a custom configuration.
To be honest, I wasn't as impressed as I wanted to be. Given what Johns described as "a huge investment in Web services at the application level" within HP's software divisions, I wanted to hear him say that all of the storage and server components, all of the software value-add features and functions, were also shortly to be Web-services enabled. The W3C standards could then be leveraged to inventory, monitor, and manage all of these HP resources, and any Web services-enabled building blocks from other vendors, in a common way. That would set the stage for me to allocate and de-allocate capacity as well as all functions in my infrastructure in a manner best suited to the data, the application, and the business process.
To his credit, and perhaps because he only came over to HP StorageWorks from HP software 18 months ago, Johns understood exactly what I meant -- which is not to say that HP has such an architecture ready for delivery today. The best he could offer was to "watch this space." He said that "a flurry of announcements will be coming from HP in this area this year." I can hardly wait.
Until then, I am pleased to report some interesting developments in my preferred interpretation of "purpose-built" storage coming from Silicon Graphics (SGI) and Novell, two companies that have been rather quiet of late but whose latest product initiatives hold much promise to propel them back to front page news in these difficult economic times. I will share the details in my next column.
For now, my advice to resellers is to spend more time helping their customers to spec out what they want their infrastructure to look like over the next few years. One obvious requirement in most shops in an infrastructure management capability that will help keep labor requirements sparse even as capacity increases. Another is a smart tape strategy, which, when combined with good archiving, can constrain the need for consumers to buy more gear. You may not make as much from the sale of tape media and software as you do from shiny new boxes of value-add disk, but you will be doing your customers a much greater service.
More next week. Until then, feel free to contact me with your views: [email protected].