theCUBE Live At Hadoop World 2014

Bucking the notion that Hadoop has to be run on off-the-shelf commodity servers, supercomputer maker Cray, Inc. surprised attendees at the Strata Conference + Hadoop World 2014 summit with the introduction of an integrated high-performance computing appliance built specifically to run the batch analytics framework.

The launch of the Urika-XA follows months of testing with several long-time customers including the Department of Energy’s Oak Ridge National Laboratory, according to Cray. The platform consists of a standard 42U rack (about the size of a refrigerator) that houses up to 48 Intel Xeon-powered servers packing as much as 1,500 cores between them. Each node comes equipped with a speedy 800-gigabyte PCIe flash card for a total of more than 38 terabytes of solid-state memory assuming a full configuration. Infiniband connectivity is also supported to enable users to take full advantage of performance afforded by the SSDs.

Coupled with the six terabytes of DRAM included in the chassis, the Urika-XA is particularly suited for running Apache Spark, a query processing component for Hadoop that is up to 100 times faster in-memory and 10 times faster on disk than the default data crunching engine. Spark is one of the software components that comes pre-installed in the appliance along with a proprietary run-time module designed to speed up analytics even further, as well as the Lustre file system.

In particular, Cray decided to equip Urika-XA with the recently upgraded distribution from Cloudera Inc. But the supercomputing giant didn’t rule out the possibility of supporting other distributions.

Rounding out the Urika-XA is a beefy Sonexion 900 disk array based on technology from Seagate Technology PLC that brings the base storage to a formidable 200 terabytes. Cray is pegging the system as a more convenient and streamlined alternative to setting up a Hadoop cluster from scratch, which is the same value proposition Oracle is touting with its similarly pre-integrated and InfiniBand-supporting BDA appliance.

The Urika-XA, which joins Cray’s existing Urike-GD graph analytics appliance, will become available for order in December. The platform is geared towards data-intensive research in bleeding edge areas such generics, financial services and materials science, with the Oak Ridge National Laboratory planning to harness it for discovering new nano-structures, according to resident data systems architect Galen Shipman.