From novel algorithms that are an order of magnitude more efficient than state of the art to extremely dense and power-efficient water cooled microservers: Dome shows paths towards exa-scale in radio astronomy.
Published by the editorial team, 6 June 2017
On May 18-19, after five years of research and development at the ASTRON & IBM centre for Exascale technology, the Dome project presented the latest results on the paths towards exascale in data transport and computing. Indeed, the technological challenges for the Square Kilometre Array (SKA) are big, and naïve scaling of conventional technologies would yield a telescope requiring in the order of a gigaWatt to run. Moreover, Moore’s law of increasing complexity and number of transistors on a chip is flattening, and it will not be able to completely bridge the gap between what is needed and what is possible.
The Dome project therefore looked and found ways to boost efficiency in three target fields: green computing that addresses technologies to radically reduce the power needed to do computationally intensive work on extremely large amounts of data, Nano-photonics that addresses technologies needed to drastically reduce the power of data transport over longer distances and inside computing machines, and Data & Streaming that addresses technologies to process data on-the-fly and store data at a high efficiency for later use.
With well over one hundred publications, several tens of SKA design reports produced, tens of architectures proposed in the compute and photonics domains, modelling tooling developed for computing architectures and for storage, hardware demonstrators created for microservers, data transport and for photonic beamforming, and supporting an active Users Platform with thirty members, the Dome project can truly be considered a success.
Scientific highlights of Dome were presented by giving an overview of the designs and of the data and computing challenges posed by the SKA, and subsequently zooming in into specific areas where Dome contributed and where Dome proposed designs that significantly improve efficiency.For example, a holistic design approach for computing architectures was presented, allowing extremely fast design space exploration. Similarly, a tiered-storage design approach with tape, disk and solid state memory was presented that automatically gives an optimal design in terms of access times and data volumes for a given cost. Various accelerator designs were presented that can run algorithms very efficiently. On the algorithm side of things, an overview was presented of the entire SKA signal processing chain, with novel efficient approaches in beamforming, calibration and imaging. The scientific Dome presentations concluded with a session on science data centres: a view from both the astronomer-user perspective, and from a technological perspective.
Secretary General Maarten Camps of the Ministry of Economic affairs noticed similarities between our approach to tackle the exascale challenge and the Ministries’ approach to tackle societal challenges. Both are based on a three-stages approach: define a goal at the horizon, define the steps (innovation) to get there, and find partners and join forces. Indeed, our “dot” at the horizon is the SKA and its regional science data centres as explained by Carole Jackson. The innovation is achieved by initiatives such as Dome, and through its collaborations.The impact of Dome on the SKA and society was explained by Gert Kruithof. For the SKA, Dome results were made available for phase one, including efficient calibration and imaging algorithms (for example accelerated gridding) and photonic links. Some results target more towards the second phase of the SKA.
Now, after five years of hard work, the Dome project comes to an end in October of this year. Dome has brought us much closer to our dot on the horizon. We're not there yet, but the symposium clearly showed next steps how to get there.