Designed to Allow for Failures
Unlike with other x86 servers within IBM's System x portfolio, the company deemphasized high availability and fault tolerance with the iDataPlex. McGee said in Web 2.0 environments redundancy is typically handled by the software running the environment. Joe Clabby, president of Clabby Analytics, said this aspect of the new system is probably the most innovative part of the design as it will allow customers to easily swap components in and out of the iDataPlex, while helping to reduce the overall cost.For now, the only other system vendors embracing this type of design are Verari Systems and Rackable Systems. But Clabby said he believes that other OEMs will follow IBM's lead if it proves successful. The difference between these smaller players and IBM right now is that IBM has the ability to tap a worldwide distribution network and its manufacturing facilities in China. "IBM has just validated their markets," Clabby wrote. Besides working for Web 2.0 companies, McGee said the iDataPlex can also work in the new types of HPC environments that financial-services enterprises and other businesses are beginning to explore in order to harness more processing power without investing in a more traditional supercomputer. When the iDataPlex System officially goes on sale June 2, IBM will offer a combination of 22 different configurations with a choice of 1U (1.75-inch) or 2U (3.5-inch) x86-based servers and a choice of switches from Cisco Systems, QLogic and others. Blade Network Technologies built a new Ethernet switch, the RackSwitch G8000, specifically for the system. All of the servers that IBM will use within the iDataPlex rack are based on Intel's "Bensley" dual-socket platform, which supports a number of different dual-core Xeon processors from both the 65-nanometer and 45-nanometer families. McGee said the iDataPlex also works with any storage array that supports IBM's System x servers. The new system also offers customers a choice between Novell and Red Hat Linux and supports xCat (Extreme Cluster Administration Toolkit), an open-source cluster management tool.
"This architecture has been designed to allow for failures-and fix-ups on the fly," Clabby wrote in an e-mail. "This is a big departure from the constant focus on high availability, fault tolerance, disaster recovery and all the other must-haves for enterprise computing environments. Taking out redundant parts reduces development and testing costs-you don't have to over-design for business resiliency-and it also helps significantly reduce system cost."