Big Data processing has usually been done by distributing workloads over lots of commodity machines, but Cray threw that notion out with a new monster system (it calls it an appliance but looks like a server rack) called Urika-XA. This is the second in the Urika line; its little brother, the Urika-GD, which offers real-time data discovery using graph-based analytics. The XA stands for "extreme analytics" because that's what it does. It is optimized around compute- and memory-intensive workloads.
A maxed out Urika-XA comes with more than 1,500 cores (48 compute nodes in a standard 42u rack case), 6TB of DRAM, 200TB of SSD flash and 120TB of disk storage, along with Infiniband to tie it all together. It runs the Lustre parallel file system, offers compatibility with the Hadoop Distributed File System (HDFS) and has POSIX compliance. It comes with Apache Hadoop and Apache Spark frameworks for Big Data analytics, along with Cray's Adaptive Runtime for Hadoop.
This all promises a turnkey solution for Big Data and analytics in general. A Big Data deployment now means buying the hardware from multiple vendors, then getting the packages together. Some of the larger players like HP and Dell are offering packaged, basic Big Data systems but they tend to be built around Hadoop, and Hadoop has its shortcomings.
Analytics is becoming more important to businesses as they look to shorten their reaction times to trends and patterns, and the promise of real-time analytics is something every retailer would welcome so they can respond immediately, not days after the fact. Cray still has to disclose more information on the kinds of workloads the system can handle and how much data the system can chew through per second.
Cray plans to make the Urika-XA available this coming December.