IT PEOPLE WHO get burned by an emerging technology rarely submit themselves to more of the same. But for one survivor...
of a virtualization project gone wrong, the benefits of the technology were still compelling enough to warrant giving it another shot.
Approximately three years ago, Todd Wyman, a senior Unix administrator at a Midwestern manufacturing company, began virtualizing his firm's 30TB of Hitachi and Hewlett-Packard disk using DataCore software. Approximately one year into the project, the environment became "unstable." While Wyman and DataCore differ on the causes of the instability (Wyman blames DataCore's in-band architecture, while DataCore claims Wyman's company hadn't properly configured its redundancy), things were bad enough that the DataCore servers were yanked.
Wyman and his co-workers worked for about six weeks to restore the data center to its original, non-virtualized state. Shortly thereafter, they started looking for another virtualization platform. "We missed things like single pane-of-glass management, being able to use open-source disk, the snapshotting, etc.," says Wyman.
After a nine-month evaluation, the company settled on StoreAge Networking Technologies' Storage Virtualization Manager (SVM). It isn't as "feature-rich" as FalconStor Software's IPStor, the other product the company evaluated, Wyman says, but they were more comfortable with its architecture, in which the virtualization server sits out-of-band, but actual data still travels directly between the server and the storage device. Wyman and his colleagues use StoreAge's multiMirror to perform replication to a disaster recovery site, and make heavy use of multiCopy, a snapshot implementation, particularly for making backup or test copies of Oracle databases.
Even today, the potential failure of a virtualization system is still a huge concern for storage administrators. "I'm an old Unix guy; I had enough trouble putting my data on a SAN, but then you ask me to put a Windows box in the middle ... It was a horse pill to swallow," says John Parrish, associate VP in charge of terminal technology at the Dallas/ Fort Worth International Airport Board, which uses DataCore to replicate application data between two coproduction sites on the airport campus. Data is continuously updated from a variety of feeds, and powers apps like flight information terminals and baggage handling. "If it went down, I'd be looking for another job," he says.
Parrish only considered in-band virtualization because the replication features he needed weren't available for his midrange Hitachi arrays, and took the plunge only after extensive in-house testing. "Knock wood, it has not faltered once," he says.
Slowly, but surely, people's concerns about virtualization are being allayed, says George Teixeira, DataCore's president and CEO. When it comes to reliability, the same rules that apply to everyone else need to be followed. "If you want high availability, you need two pipes to everything," he says. If you skimp and, for example, use only a single controller and it fails, "you're in deep yogurt."