Blades cool hot, crowded data center

Tier 1 suppliers aren't the only ones that can come up with highly efficient Linux clusters. Just ask the managment team at Silicon Valley's Magma Design Automation.

We moved from Solaris to Linux for performance, not cost reasons.
Abel Chow
senior Unix administratorMagma Design Automation

The hundreds of servers that made up Magma Design Automation's Linux-based clustering environment took up too much space, ran too hot and were difficult to manage from a central location.

When you're in the business of providing hosted software to some of the biggest semi-conductor companies in the world, the efficiency of your server cluster is of paramount importance. It was apparent to Magma's executives that the time to make a change had come.

Magma's data center had been overflowing with a huge cluster of Intel-based, duel processor rack-mounted servers. But as the company's data center scaled over time, so did its server cooling and cluster management issues.

The company wanted to replace many of its servers with smaller blade servers that ran on Opteron processors, which tend to run cooler and faster than their Intel counterparts.

After much consideration, the company decided to skip over tier one vendors like IBM and Sun Microsystems and purchase a clustering environment from a smaller local vendor. The company chose Appro International's Appro HyperBlade Cluster Solution.

"One of the primary reasons for going with Appro versus a tier one vendor was that they had the latest [and fastest] technology a couple of months before other vendors," said Abel Chow, Magma's senior Unix administrator. "Cost [was] also a big factor."

MORE ON THIS TOPIC:

Special report: Linux clusters get supercharged

Appro to sell thin Opteron 64-bit HPC servers

Increased space, remote management capabilities seen as key

Chow said the time from delivery of the new systems to production was reduced significantly by having Appro handle setup and installation. Today, more than half of the company's servers are Appro blades.

"With a normal rack-mount system, you could put 40 in a rack," Chow said. "With the blades, we can put 80 in a rack. The blade racks are a little wider, but still get better density."

Even more appealing to Chow was the fact that the blade servers could be more easily managed through Appro's BladeDome remote management software. Chow uses the software to connect to an individual server's serial console remotely and turn the power on or off, or reset the machine as needed.

Prior to implementing the remote management software, Chow had to walk around to each server individually by attaching a monitor and keyboard and searching for the problem.

"Having a single point of management with the BladeDome simplifies monitoring and maintenance of the machines in the cluster," Chow said. "Being able to see a machine having a trouble condition at a glance allows us to be more responsive to problems with the machines."

Linux for performance?

Chow pointed out that the operating systems running in his data center include Red Hat Enterprise Linux 3, Red Hat Linux 7.2 and SuSE Linux. A few years ago, the company was running Sun's Solaris, a flavor of Unix, but decided to switch to Linux for reason people don't often hear.

"We moved from Solaris to Linux for performance, not cost," Chow said. "The highest performing CPUs out there run Linux."

A satisfied customer

All told, Chow said Magma has been very pleased with its new clustering solution. Appro has been a major advantage in terms of their support needs.

"As we've purchased more of these blade servers, Appro has made small changes in design to address the concerns we've had in terms of routing the cables [and adding] more integrated features, so that I don't have to plug in another cable," Chow said.

Dig deeper on Linux high-performance computing and supercomputing

0 comments

Oldest 

Forgot Password?

No problem! Submit your e-mail address below. We'll send you an email containing your password.

Your password has been sent to:

-ADS BY GOOGLE

SearchDataCenter

SearchServerVirtualization

SearchCloudComputing

SearchEnterpriseDesktop

Close