In-Depth

Getting to Success in Storage Management

Open-source initiatives may not be workable—or even enough. Consider treating disk as inventory instead.

• By Jon William Toigo
• 11/15/2005

In a word, heterogeneous storage management is a mess, and it doesn’t show signs of getting much better any time soon—though there are some developments worth monitoring.

Let’s begin at the beginning. The need for effective storage management is self evident, both as a proactive technique for identifying and resolving capacity allocation issues and avoiding downtime, and as a means for troubleshooting application performance issues that increasingly track back to data layout and hosting.

Everybody gets it. Managing storage is necessary to do more with less and realize any sort of business value from the storage investments that currently account for about 40 cents of every dollar being spent on IT hardware today.

Storage management isn’t data management or information lifecycle management, of course, though the boundaries between the two disciplines are often as blurry as the distant horizon on a hot day. Data management looks at capacity utilization efficiency—whether the right data is parked on the right platform based on access and update characteristics, considerations such as retention and audit requirements, and platform performance/cost factors. Data management is not about allocation efficiency per se.

There is a need to interface storage management with data management to realize a soup-to-nuts data management solution, but that, too, is a different issue.

Storage management—managing capacity allocation and keeping storage infrastructure up and running—is important in these financially challenging times. In recognition of this fact, the market is saturated with storage resource management products, backup management products, archive management products, virtualization wares, data movers, and a myriad of other software components that the consumer is left to cobble together if there is any hope of making sense and purpose of any storage infrastructure.

Hurdles to Building Your Own Solution

Alternatives to rolling your own storage management solution are to (1) buy an “SRM suite” from a software vendor or (2) buy the software stack from a single hardware vendor that only works well on that vendor’s platforms or products (go homogeneous).

The first alternative confronts several hurdles, the first of which is integration. How integrated are the tools listed on the glossy vendor brochure? Do they share the same code base, or backend database, or metadata repository? Or are they integrated in the sense that they all launch from the same window, or even just because all of the components are listed in the same brochure (the expression for this is “brochure-level integration”)? If the vendor is selling integrated wares, you owe it to yourself to find out what they mean by the term.

The second hurdle is platform support. Be careful as you explore the meaning of “heterogeneous” in product descriptions. Does heterogeneous platform support mean server operating system support (e.g., Linux, UNIX, Windows, and some or all variants thereof)? Does it mean support for a variety of array products from different brand vendors? Does it mean support for different products within a given array vendor’s product family? These are all important questions to be answered if you want to avoid “gotchas” later on.

Support claims also need to be validated in terms of topology. When a vendor says it supports server OS XYZ, does that extend to clustered versions of the OS environment? Does the SRM toolset see storage through the eyes of each server or only through the nominal cluster manager (if there is one)? On the storage side, does the product support devices in a Fibre Channel fabric? What about devices connected with TCP/IP or other networks?

You'll need to determine whether the product offers visibility into storage at the LUN level (what a server sees as a volume) or does it give visibility into the actual physical disks in the array? Most products offer a mixture of both, reflecting the real stumbling block in SRM today: lack of visibility into array hardware.

A number of the large-array manufacturers don’t provide visibility into their physical hardware. A vendor might argue that this is to shield the consumer against “complexity.” Their rationale: you don’t need to know all of the engineering details about your car in order to drive it or to know when you need to refill the gasoline tank. Sounds reasonable enough.

The difference, however, is that all automobiles do provide information about the miles per gallon (or kilometers per liter) the vehicle is supposedly getting. Most storage vendors stop well short of giving you that kind of information about their products. Simply put, they don’t feel that you need to know how much of the formatted storage capacity for which you paid potentially big money is actually reserved for use by the array operating system and “value-add software.”

Nor are they likely to give an application programming interface (API) or other mechanism to a third-party management software vendor to facilitate the collection of that kind of information. That is a key hurdle for SRM. Software developers must go hat in hand to each hardware vendor and beg for access to whatever APIs or SNMP MIBs or other hooks that the hardware vendor has built into his product if they want to give consumers any value at all with their SRM products.

Even if you ask an SRM suite vendor whether they support all of the hardware directly that you have deployed on the floor (not indirectly through exposed LUNs), their answer is probably meaningless. Support for this year’s API may have been acquired from vendor XYZ. However, with the release of next year’s product, that access will need to be renegotiated. If the software representative is no longer on the hardware vendor's Christmas card list, all bets are off.

Hardware vendors are also reluctant to certify any management tools for use with their products that might inhibit or compete with the sales of its own software tools. No hardware vendor, for example, wants to make it easy to migrate data from their frame and onto a competitor’s frame, especially if they have their own tools for migrating data between two of their own products. So, whatever cross-platform functionality is offered by the SRM vendor, it's likely to be uneven: some products in the suite support some platforms while others support a different subset.

New Initiatives

The situation described above has existed since the dawn of “open systems” and is unlikely to go away any time soon. The Storage Networking Industry Association tried to define a middleware layer that would enable all hardware to connect in a management metamodel: the so-called Storage Management Initiative - Specification (SMI-S). For all the public pronouncements of support by leading vendors, the outcome has not solved the underlying problems.

Several weeks ago, IBM and a group of others announced Aperi, an open-source initiative to fix the problems. The effort might have wheels, though a majority of consumers I’ve talked to aren’t betting the ranch. Notably absent from the discussion are Microsoft, EMC, HP/AppIQ, HDS, and a preponderance of smaller vendors who can barely afford their SNIA dues, let alone join another initiative.

Of course, a true open-source initiative doesn’t require the cooperation of any vendors. All you need is a bunch of disgruntled storage managers who have come up with their own workarounds to the proprietary barriers to platform visibility and who are willing to share them in a common code pool. Maybe, in the final analysis, that is what really needs to happen.

Treat Disk as Inventory

Something else needed is a willingness among consumers to think outside industry-speak with respect to storage management. Start treating disk as inventory and manage it accordingly, using time-honored methods from the era before the so-called Information Age. When inventory is stale, get rid of it. When demand is high, stock more; but, never buy more than you need because today’s disk is getting cheaper and more capacious by the minute.

How do you evaluate inventory? Simple—by the use that is made of it. Effective storage management in the final analysis depends on effective data management. Understanding what your applications and their data require of storage services is the best way to provision infrastructure. It’s an old idea that got shelved when vendors tried to sell us one-size-fits-most infrastructure. We need to resurrect it now.

Your comments are welcome. Write to [email protected]