The Next Frontier: Data Virtualization
Executive Editor Tom Valovic
and I took a briefing this morning with virtualization vendor xkoto
. Xkoto does data virtualization focused around databases, offering (according to the company) better availability and performance, even with clustering solutions.
Its product is called Gridscale, a software appliance that works by changing the traditional active/passive roles of cluster servers and making every server active, resulting in an active/active environment. Normally, the passive servers in a cluster are hibernating, waiting for a failure to wake up and start working. Not so with Gridscale, which keeps all servers active and humming along. The company claims up to an 85 percent performance improvement with this arrangement, depending on the application and environment, of course.
Those are lofty figures; if you've used Gridscale, please let me know what kind of performance gains you're seeing.
Gridscale virtualizes the data infrastructure (it sits between the app and the database), so it doesn't matter whether the data is stored locally, on the LAN or in Timbuktu. It load balances the requests, ensuring consistent access to data. Since all the clustered servers are active, it also means that a server can be taken offline at any time for patching or maintenance, without affecting data access.
Another nice feature of Gridscale is that, unlike some clustered arrangements, it doesn't have to be matching hardware. You can use your older servers, your newer servers -- basically any server. You don't need a complete matching set of hardware to run it on.
Currently, Gridscale supports IBM DB2 on various Windows, Linux, Sun and IBM operating systems. Officials told us that Microsoft SQL Server, currently in closed beta testing, will be supported "later this year." Xkoto has worked very closely with IBM, and is developing a similar relationship with Microsoft, officials said in the briefing.
One thing that intrigues me about this technology is the potential for virtualizing databases. As you may know, many admins are loathe to virtualize apps like databases. Their high I/O requirements normally don't make them good candidates, given the processing overhead virtualization adds. However, if there are significant database performance gains through use of Gridscale, that might overcome those well-founded fears.
If you're a DBA or otherwise have responsibility for data management or virtualization, is this something that might convince you to virtualize your databases? I'd be very interested to hear from you on this issue.
Posted by Keith Ward on 04/25/2008 at 12:48 PM