4 Lessons Learned From Virtualization Masters

IT pros who've virtualized upwards of 70% of the data center share how they've solved problems that more cautious adopters have yet to see.
My Mistake: 10 CIOs Share Do-Over Worthy Moments
Slideshow: My Mistake: 10 CIOs Share Do-Over Worthy Moments
(click image for larger view and for slideshow)
Lesson 3: Transaction databases are fair game for virtual servers.

It's been an unwritten rule of virtualization to keep hands off the database system. It was too mission critical and its operation too unpredictable in the more complex virtualized environment, where calls for hardware services have to be passed through a hypervisor. But Accenture's Jay Corn, lead North American consultant for infrastructure and consolidation, says Accenture has done it many times and the inhibitions over databases running inside a virtual machine are misplaced.

Oracle has a provision in its support contract that says if a problem occurs with its database running inside another vendor's virtual machine product, it may need to be duplicated outside the virtual machine before Oracle technical staff troubleshoots the issue. In the advanced implementer's experience, that qualification has rarely been invoked, they said. Nevertheless, its existence probably has something to do with those previously mentioned inhibitions.

"We have successfully virtualized Oracle database servers many times. I can't think of a case where we've had a problem," said Corn in an interview. That is, database functions and data handling operate the same way, with no data loss, in the virtualized environment as well as physical.

Lesson 4: Choose a high-speed highway or one with more lanes.

Aaron Branham, director of IT at Bluelock, an infrastructure as a service provider in Indianapolis, says there's a difference for virtualized environments between Intel's latest chips, found in Cisco's UCS, and an environment based on AMD servers.

Intel Xeon chips in UCS will do more work faster, but it's not as easy to spread a broad set of workloads over them, said Branham, who considered the UCS architecture as the basis for his cloud service. The AMD chips don't run quite so fast, but come with up to 12 cores included. That makes it easier to spread a variety of workloads over the server, with each load getting its own core.

"Cisco did unique things with the memory in UCS," said Branham. (It put an ASIC chip in the UCS server that serves as a memory multiplier, raising the RAM limit.) That would be highly useful for a large, data-intensive workload where response times were important; more data could reside in memory. Branham, on the other hand, needed to run a large variety of small workloads for his customers, and he put together a combination of AMD servers and the Xsigo server that virtualizes I/O for virtualized hosts, Xsigo Director.

"With Intel (and UCS), you get a four-lane highway with a 70-mph speed limit. With AMD, you get a six-lane highway with a 55-mph limit," he said.

Charles Babcock is an editor-at-large for InformationWeek.

Editor's Choice
Brian T. Horowitz, Contributing Reporter
Samuel Greengard, Contributing Reporter
Nathan Eddy, Freelance Writer
Brandon Taylor, Digital Editorial Program Manager
Jessica Davis, Senior Editor
Cynthia Harvey, Freelance Journalist, InformationWeek
Sara Peters, Editor-in-Chief, InformationWeek / Network Computing