>> continued from previous page
Beyond the Virtual
All volume-virtualization schemes depend on device drivers that eliminate operating system controls. In-band and out-of-band techniques substitute a unique driver for the storage arrays' own storage device drivers. Host-based approaches actually create a device driver as a function of software running on the application host.
Host-based approaches also depend on secondary software components to perform end runs that spoof the operating system to help it understand that a volume that was 32 GB yesterday is 64 GB today. In performing essentially illegal operating system calls, such volume resizing utilities are prone to failure despite vendor claims to the contrary.
Eliminating operating system control over volumes is a prerequisite to realizing contemporary SANs' scalability objectives. However, virtualization's value proposition does not stop with the delivery of a successful abstraction layer.
Vendors have suggested that virtualization will reduce storage-management costs by reducing the number of storage targets that must be managed and the effort required to manage them. In fact, virtualization has no inherent reduced cost of management value; it simply provides a layer of software functionality that can be leveraged in a number of ways to aid in managing storage.
This point is key to understanding the state of the virtualization industry. Vendors are adding value to their products by bundling software, unrelated to virtualization, that delivers additional storage-management capabilities.
For example, by wedding a proprietary device polling scheme to a virtualization engine, DataCore provides a method for in-band device monitoring over a Fibre Channel fabric--a service that the Fibre Channel protocol does not support natively. Using this feature, IT managers can monitor devices deployed in a Fibre Channel fabric without deploying a secondary Ethernet network to interconnect every device for monitoring and management.
Similarly, if virtualization-software vendors develop policy-based storage resource allocation software that takes advantage of their virtualization software foundation, they can make good on claims of optimizing storage utilization efficiency. Fujitsu Softek, which bundles DataCore's virtualization software into its own policy-based management tools, is a leader in this evolving space. Theoretically, a policy-based provisioning system could interact with the virtualization engine to adapt volume size to application requirements on the fly. Given IT's limited acceptance of policy-based management up to now, however, linking the two might well be the kiss of death for both technologies.
Additional software used in conjunction with a virtualization engine could also deliver fault tolerance. Such software could enable storage resources to be taken out of service for maintenance after automatically replicating its data to other storage resources so as not to disturb volume integrity.
Virtualization alone delivers none of these benefits. This is more than an academic point. Much of the mistrust of vendors' claims has to do with the hype about virtualization's benefits. Virtualization itself has little or no business value. Additional software is required to realize vendor ROI claims.
The Dark Side
In fact, virtualization software can actually impair the efficiency of certain storage-management applications, such as data restores from tape backups (see "Storage Disaster: Will You Recover?"). If the backup-and-restore software used is not virtualization-engine-aware, writing files from tape back to volumes produced by a virtualization engine can take an extraordinary amount of time (100 hours to restore a terabyte of data, for example). Why? The virtualization engine acts as a write filter (see "Potential Choke Point in Tape-Based Data Restore"). Writing data to a virtual volume introduces latency as the engine works to target incoming data to the physical devices that make up the volume. Large data movements exacerbate the problem.
A similar issue, known as the "write penalty," existed in early software-based RAID arrays. Engineers took more than a decade to finesse the problem, by moving RAID software into faster executing array controller hardware and adding memory caches. Virtualization software is so new that write penalties have yet to be addressed--or even encountered. Storage software vendors will need to do extensive testing to ensure compatibility with the virtualization products that obtain industry hegemony.
The Bottom Line
Storage virtualization remains a bleeding-edge technology. There are no standards to describe how it should be implemented or how, as a service, it should be handled or leveraged by other storage-management software products.
Most IT managers are wise to these facts, according to Kerns--they know the problem revolves around standardization and vendor pedigree. Virtualization's market potential is "still there and huge, but only if large enterprises buy into it," he says. Fortune 100 companies, he notes, have been hesitant to deploy virtualization solutions because those organizations "buy only from big-name vendors. All the current virtualization vendors were unknown companies until [Hewlett-Packard] acquired StorageApps last year." Products from that acquisition, and from Compaq, are just beginning to roll out.
Although virtualization may be inevitable as a prerequisite for truly networked storage, IT decision makers are confident one of the products will provide the industry-dominant virtualization technique. Until then, they will stick to the virtualization methods they know best: those provided by file systems and array controllers.
Jon William Toigo is a frequent contributor to Network Computing and has written 12 books, including The Holy Grail of Networked Storage Management, due out later this year from Prentice Hall PTR. Send your comments on this article to him at jtoigo@intnet.net.