In the Absence of a Data-Centric Storage Management Solution

The cursory description of a data-centric management scheme provided in the preceding pages is offered in the hopes that it will stimulate increased interest in and development of technology based on the concept of self-describing data. In and of itself, a data-naming scheme does not effect change in storage capacity utilization efficiency. If it is sufficiently open and compatible with existing operating systems, applications and storage platforms, it will enable the development of improved processes for managing data.

In the final analysis, a data-centric management approach makes sense. If we agree that storage is fast becoming its own infrastructure, whether as a result of increasingly networked topologies or as a consequence of the evolutionary deconstruction of the von Neumann machine, we must also agree that storage is no longer a repository, a place where data goes to rest. Thus management techniques focused on the maintenance of disk arrays is no longer valid.

Today, the entire storage infrastructure, including switches and cables, bridge/routers and array caches, disk and tape media, host bus adapters, and system caches, is better viewed as a dynamic system. Within this dynamic system, data is constantly in motion. We understand the more that we look at capacity utilization inefficiency, which has its roots in the mismatching of the frequency of access characteristics of data with the capabilities and expense of storage platforms. This inefficiency is expressed monetarily as the cost incurred by storing your least-often accessed data on your most expensive platforms ”expensive because they are designed to provide the fastest data access.

References to "speed of access" imply motion, and references to "frequency of access" imply time. As these expressions creep more and more into our descriptions of the storage infrastructure within our shops , it should remind us that storage is not a static repository, but a dynamic system that requires a management philosophy and capability capable of coping with points, planes, objects, and time.

What should be done until self-describing data opens the door for capacity utilization efficiency improvements? That is a difficult question, like asking what should be done until the storage infrastructure becomes an intelligent utility.

Practically speaking, we must do what we have always done. Provision storage as efficiently as our wits and the blunt tools available for management allow.

Many storage management software vendors are working in earnest to devise a kind of SMS capability for open systems storage as this book goes to press. Most are leveraging virtualization technology, such as it is, to help define "Storage Groups." They are also leveraging policy engines such as those from CreekPath, together with SRM tools, and automated storage migration and replication technologies, to help define "Storage Classes." Their goal is to remove some of the manual effort that goes into provisioning capacity to applications.

The selection of which storage management architecture to select is a judgement call. Take care not to be swayed by vendor hype. The option always exists to purchase hardware only from a single vendor or vendor cadre. This approach offers ease of management since that vendor's point solution may be all you need to manage your environment.

This solution is generally less preferable in settings where an investment has been made in heterogeneous infrastructure. It also has the drawback of introducing dependency on a single vendor that may not be desirable for a number of valid reasons. For one, you are always vulnerable to sacrificing advantageous pricing or best-of-breed technology in favor of convenience.

Still, this approach can establish a stable platform ” assuming the stability of the vendor ”that offers certain management conveniences ”not unlike the mainframe data center of old.

Another option is to purchase only the newest breed of hardware platforms from a growing number of vendors who have engineered " self-management " features into their hardware. Vendors ranging from Hewlett-Packard to IBM to EMC are preparing to address the problem of management burden by releasing to market hardware that is augmented with resident automated management utilities that "do it all so you don't have to."

This is essentially the SAN vision done another way: Instead of some SAN software providing application intelligent provisioning and self maintenance, the devices in the storage topology are themselves intelligent and deliver allocation efficiency to the applications that they serve. If the products live up to their claims, if they are truly "configure and forget" and will "phone home" to have themselves replaced at the first sign of trouble, they will certainly alleviate the burden upon (and the need for) storage managers.

The problem at this point is that vendor claims cannot be validated because products have not yet shipped. Moreover, there is no certainty whatsoever that products from different vendors will plug and play into the same infrastructure. Without further testing, it is impossible to tell whether these intelligent peripherals will provide any better efficiency than the proprietary boxes that they replace. You have only the vendor's assurances, which may be pretty difficult to believe given the existence of so much non-interoperable technology in the market today.

Option three is to manage storage capacity allocation and utilization the "old-fashioned way." Identify software products that seem to fix immediate problems ”preferably ones with forward-looking roadmaps that give you confidence about the product (and the vendor's) longevity. Don't be afraid to consider products from newer vendors, many of whom have more innovative technology than name brand players (in part, because they have no ties to older approaches or an existing install base), especially when these products can deliver an immediate return on investment.

Needless to say, there is a lot of push back from many companies to this recommendation. Many IT managers have become conservative about the technology they are willing to try and believe there is a certain safety in buying products only from companies with three-letter acronyms for names (e.g., IBM, EMC, etc.) BMC Software demonstrated the fallacy inherent in this view in 2003. The vendor, a company with brand-name recognition and a sterling reputation, simply discontinued its application-centric storage management product and stranded 130 customers. ^[12]

There is really no reason not to be open to new technologies. The worst that could happen is that your storage will continue to be managed the same way that it is today: with all of the inefficiencies of a still-maturing technology segment. At best, you might actually develop a more cost-effective storage infrastructure.

Finally, if you find a software mix that works, make it a key determinant of the hardware you buy going forward. Tell your vendors that you won't buy their gear unless it supports the management approach you have determined to be the best for your shop. You would be surprised how quickly vendors can respond to requests for improved management support when confronted by enough customers who demand it.

Hardware selection also needs to be guided by a pragmatic assessment of application requirements and data access characteristics. Storage managers must identify these requirements proactively and realistically in order to make intelligent choices about storage platforms. Don't deploy a SAN when a NAS or direct-attached array with a scalable back-end delivers the same capability with less hardware to manage.

One last criterion in storage infrastructure hardware and software selection is data protection. Next to storage provisioning, data protection is the second most painful activity cited by storage management professionals. It is to this subject that we turn our attention in Chapter 9.