Aus defence on hunt for $2.2M supercomputer

The Australian Defence Science and Technology Organisation is in the market for a supercomputer to solve massive computational fluid simulations, and has gone as far as trading power redundancy for even more computing power.

The Defence Science and Technology Organisation (DSTO) is searching for a new supercomputer to help it solve fluid dynamics problems with as little hands-on requirements as possible.

Going to market with a request for tender on Friday, DSTO said that it would use the high-performance computer (HPC) cluster to run simulations, typically over OpenFOAM, an open-source fluid flows toolbox that has been used in the past to solve or run simulations not only for classical fluid flow, but also for chemical reactions and heat transfers. DSTO's specification states that the procurement of hardware, software, and services is a one-off contract, and that it is willing to spend up to AU$2.2 million.

The HPC is expected to be installed at DSTO's Melbourne facility, providing context into what it will ultimately be used for. Major projects at DSTO Melbourne include research into aerospace, surface, undersea, and land platforms; protecting combatants from chemical, biological, and radiological environments; and propulsion mechanics.

While DSTO has stated that it wants the cluster to be able to perform small and large tasks, it expects to use it for one or two large-scale jobs the majority of the time.

Australia's defence department is well known for its need to create custom software, rather than use off-the-shelf products that could introduce supply chain risks. However, for this tender, DSTO wants as hands-off an approach as possible. It has stated that it will "not be in a position to modify source code or invest time to resolve potential future software incompatibilities with such hardware," and, as such, "the process of performance-tuning software and hardware, to ensure the system is operating with optimal settings, should be straightforward and documented."

Processors in the cluster are expected to be x86-64 based, and each compute node CPU will have six to eight cores running at a minimum of 2.6Ghz.

The compute nodes themselves will not be protected by UPS, with DSTO opting to trade the reduction in overhead for even higher computing performance. Due to the cluster being vulnerable to even momentary power loss, DSTO has an additional requirement such that users are notified when jobs are interrupted, and that they can be resumed once power is restored.

On the other hand, the cluster-head nodes and storage nodes must be protected by UPS. This will be provided by DSTO if the head nodes are installed directly in its datacentre, or provided by the tender if they are located elsewhere.

Storage for the HPC will come from DSTO's existing 90TB of network-attached storage (NAS).

DSTO hopes to award the contract by the beginning of March, have all hardware delivered by April, and have the HPC commissioned by May 3. The successful applicant will then enter into a five-year warranty period.