Allinea Tools to Help Professor Stephen Hawking's Consortium Get More Bang From COSMOS
Cosmology's universe expands with unified HPC development tools from Arm Software
Cambridge, UK ― Over the past two decades, cosmology has emerged as a highly competitive, data-driven field. With the success of space and ground-based experiments, scientists who once immersed themselves in the elegance of the beautiful symmetries of an equation can now test their theories against an unprecedented amount of data collected from the far reaches of the universe.
"Cosmology has become a precision science driven by new data. We are exploiting the Planck satellite data of the cosmic microwave sky, as well as billion galaxy surveys, to better understand our place in the universe," says Professor Stephen Hawking, Principal Investigator for the UK’s COSMOS Consortium. "Without supercomputers like COSMOS, we would not be able to reach out and make contact between theory and the real universe to test whether our ideas are right."
While physicists must harness the power of HPC, they do not want to get bogged down in the technical details. Which is why the COSMOS consortium, Europe’s leading group of cosmology investigators brought together by Stephen Hawking, has entered into collaboration with Arm Software. They are bringing Arm DDT and Arm MAP to COSMOS_IX, one of the world’s largest shared-memory supercomputers and the world’s first symmetric multiprocessing system to be boosted by Intel® Xeon® Phi™ coprocessors. With COSMOS_IX being a part of the DiRAC – the UK’s integrated supercomputing facility for theoretical modeling and HPC-based research – many researchers in particle physics, astronomy and cosmology will get access to the powerful Arm HPC tools.
Arm DDT and Arm MAP will help the consortium adapt its codes to the new architectures.
"Parallel programming is essential when exploring big science challenges. Intel® Xeon® Phi™ coprocessors offer a highly parallel capability without losing the familiar languages, models, and tools that software developers rely on," said James Reinders, Director and Parallel Programming Evangelist, Intel® Corporation.
"Arm DDT and Arm MAP fit this philosophy by offering profiling and debugging power in a unified, simple interface across Intel® Xeon® processors and Intel® Xeon® Phi™ coprocessors. This ease-of-use is a significant benefit for scientists who need to harness parallel programming but focus on their science."
COSMOS System Administrator Andrey Kaliazin agrees, adding that science is very much like business these days with scientists facing ongoing deadline pressure to produce results for academic conferences and funding bodies.
"Cosmology is extremely competitive with a lot of groups working on related projects so it is important to keep up and have an edge; and where everything depends on numerical simulations, you have to have your code valid, consistent and fast to get the results when you need them."
Arm DDT adapts code to analyze Planck satellite data
COSMOS parallel programmer James Briggs has been using Arm DDT to help adapt CAMB – a complex code used to analyze data from the NASA WMAP and ESA Planck satellites – for the Intel® Xeon® Phi™ architecture. Planck, successor to WMAP, was Europe’s first space mission to study relic radiation from the Big Bang. Since 1992, scientists have been using small temperature fluctuations in this radiation to understand the origin of the Universe and the formation of galaxies.
"Arm DDT is so clear. It gave a lot of insights which helped James move forward," says Andrey Kaliazin. "It gives you an overview and helps you look into things you wouldn’t expect to think of and look at."
The consortium is focused on using the Intel® Xeon® Phi™ coprocessors in offload mode, where bits of the code that are highly parallel are executed on the co-processors. James Briggs says Arm DDT could be particularly crucial to assist with the complex dance of data as it moves back and forth from the coprocessors to the main memory.
"We need to follow up with how the variables are updated on the coprocessors and the main processors. This is hard to do manually so this is where Arm DDT will come into play."
Legacy Codes Demand a Unified Tool
Another huge challenge for the consortium, and indeed all cosmologists, is the existence of legacy codes written in Fortran and C that are in various stages of parallelization.
Andrey Kaliazin says the consortium anticipates using Arm MAP in combination with Arm DDT to update these codes.
"Bringing these two tools under one hood creates a clear advantage for us," he says. "We could run Arm MAP to see where the bottlenecks are, then attack that particular part of the code to parallelize it, while keeping it bug-free with Arm DDT, then run it through the profiler again, and go in a circle until we get the whole thing parallelized as much as possible."