Virtual Domain

Blog archive

Virtual Memory

Most of us know of an older folk or two plagued by memory problems. Such is the case with x86 server virtualization these days. As the technology has aged -- and in its case, proliferated -- memory is becoming an issue.

In fact, ask enterprise IT managers what causes them the most problems as they virtualize their infrastructures and, almost unanimously, they'll reply "memory," contends Andy Mallinger, vice president of marketing for RNA Networks, Inc. Follow up with a question on how they're addressing the problem and they'll say they're adding more physical memory or servers -- costly remedies from budget and overhead perspectives.

RNA proposes a new way to deal with memory -- virtualize it. In other words, it has come up with a way to make memory a shared, network resource.

"The amount of memory available to pool depends on the servers and application, but in general, you can start with the assumption that you can make 50 percent of memory in the data center a shared resource," Mallinger says. Of course, he notes, you don't have to virtualize all memory. "You wouldn't touch the memory for critical apps and you can keep dedicated memory for some devices. But basically there's sufficient memory in the data center that's not being used."

And rest assured, you always can move memory back to a server if need be, too, he adds. For now, that process would be manual but one day RNA hopes to make it automated, he says.

RNA calls its underlying technology the Memory Virtualization Platform (MVP). On top of that platform it offers two products to date: RNAmessenger, for low-latency applications; and RNAcache, for transaction-heavy applications.
Dan Kuznetsky, vice president of research operations for The 451 Group, finds RNAcache particularly interesting for those enterprises that have extreme transaction processing needs.
The RNAcache software lets applications load their entire working dataset into a memory cache for faster access and processing, RNA describes. It says this so-called memory virtualization technology is for use by enterprises doing predictive analytics and high-volume Internet applications such as travel reservations, for example.
"This stuff is very powerful. Anyone doing a lot of transaction processing and raw analytic work could make great use of this," says Kuznetsky, noting, however, that virtualization at this level is neither easy to understand nor necessary technology for the average virtualization shop right now. "Today, only very sophisticated customer environments, like those at the high end of the financial services market, need to consume this technology."

When a company does need this technology -- watch out. Mallinger relates how one customer, a hedge fund, wanted to boost the number of transactions it processes per second from 5,000 to 10,000. Using the RNA technology, it now processes 50,000 transactions per second.

That's fast.

Posted by Beth Schultz on 09/16/2009 at 12:49 PM


Subscribe on YouTube