ARM in the datacenter gets another boost
The ongoing conversation about how ARM (and other potential low-power, massively dense packed computing systems) will be transforming the datacenter has focused primarily on the energy –efficiecy and power and cost savings that are potentially represented by moving appropriate workloads to these new environments. The forthcoming 64-bit ARMv8 technology has intensified that discussion, as the appearance f a 64-bit model will address many of the concerns of potential users in terms of flexibility and capabilities of ARM devices. But today’s announcement from Univa cuts straight to the heart of the matter on deploy9ing ARM-based servers in the datacenter.
Univa has released the beta of their Enterprise Workload Management system, the Univa Grid Engine with support for ARM v6 and v7 CPUs. The Grid Engine is a distributed resource management platform the enables the deployment of workloads such as Hadoop across computing cores throughout the processing environment. The addition of ARM to the supported CPU set means that the scheduler can now assign appropriate tasks to these systems, as well as existing x86 cores in the infrastructure. The Grid Engine already supports multiple different versions of Linux, Microsoft Windows Server, Solaris, HP-UX, and AIX as host operating systems, and as appropriate code is made available for ARM servers running, for example, Linux, users of the Grid Engine will be able to include those servers in their automated workload scheduling and assignment.
The initial ARM implementation is being supported by teaming up with Calxeda, one of the leading players in the densely packed ARM SOC datacenter server market. HP, for example, uses the Calxeda SOC in the AR-based server that they fiorst announced in late 2011, so there is an existing market of servers ale=ready in place suitable for the deployment of the grid Engine.
Much like cloud technology, the Univa Grid Engine is an enabling technology that will eventually lead to an improved ability for IT to just assign workloads to the datacenter with the underlying infrastructure no longer being a primary concern. Customers interested in the beta software can find it here.