Cloud storage: Lower cost and increase uptime

Hard on the heels of yesterday’s win for Cray at The Pawsey Centre, the Murchison widefield array in Western Australia is pulling in some new iron in the form of a high-powered Linux cluster from IBM.

As previously noted in The Register, modern radioastronomy poses a considerable computing challenge: it generates too much data to be stored in raw form. To help deal with this, processing near the instrument site is needed to thin out the data.

The array replaces Murchison’s current custom-built FPGA-based processing environment. The processor needs to integrate data from the site’s 4,096 dipole antennas, which means it will have to cope with processing 50 TB of observation data each day, at a rate of 8 GB per second.

To get a handle on the scope of the challenge: the signal correlation performed at Murchison involves combining 500,000 signal pairs, producing 3 billion “visibility points” per second. GPU-based processing has long been on the menu for the facility, which published this document in 2009 detailing its proposed approaches.

Each of the dx360s will be configured with dual Xeon processors and two NVIDIA Tesla GPUs, and will have 15 PB of local storage.

The remote Murchison widefield array is designed to sniff out lower-frequency signals from the early universe, to try and gain insights into the formation of the first stars and galaxies. As work proceeds on the Square Kilometer Array (earlier this year split between Australia and South Africa), Murchison will also host much of Australia’s SKA infrastructure. ®

Steps to Take Before Choosing a Business Continuity Partner