The great thing about working at SGI is the customers and partners we work with. One of the oldest customers of SGI is the Stephen Hawking Centre for Theoretical Cosmology (CTC), and one of our most valued technology partners is Intel.
The team at CTC are constantly looking at innovative ways to test their new theories against observational data. It’s a very competitive field, so time to publish results is key. CTC have utilized multiple generations of SGI shared memory systems to accelerate their computations and workflows, by minimizing complexity of their environment and focusing on productivity.
Most recently CTC are moving to advanced visualization to enable them to analyze extremely large datasets that have never been possible to visualize before (up to 10TB capability has now been demonstrated). This work has been carried out in conjunction with Intel and SGI, leveraging the OSPRay visualization tool (https://ospray.github.io/), along with the SGI® UV™ 2000 (1856 cores, 14.5TB memory) in use at CTC.
The Walls code has been in use for over 10 years at CTC to understand the formation of domain walls in the early universe. If you consider today is time=14 billon years, then the simulation starts at time=200,000 years (just before electrons and protons combine to form hydrogen and the cosmic microwave background (CMB) radiation is released)
Figure 1 Early Universe Domain Walls
For more background on related work, please read from the COSMOS SGI White Paper on the Walls code, entitled “Unveiling the Early Universe with Intel® Xeon® Processors and Intel® Xeon Phi™ Coprocessors”, which can be found here.
To gain better insight into the processes taking place, interactive visualization is an important tool. A key aspect is that the data generated by the simulations is held in-memory, and is available for other processes to perform in-situ analysis, such as the visualization. In fact, multiple visualizations can be performed concurrently, leveraging the same dataset in-memory.
Figure 2 Different levels of zoom applied to Walls visualization
The ability to generate new data has also been accelerated by the CTC, leveraging their skills as an IPPC (Intel Parallel Processing Center), where a focus on code modernization has enabled increased performance of the Walls code on both the Intel® Xeon® and Intel® Xeon® Phi processors in the SGI UV 2000 system.
This is a really a prelude to what will become the most productive real-time HPC discovery workflows of the future – large datasets held in-memory, being produced by simulation (accelerated modernized code), or ingested from state of the art instruments, or other data feeds, coupled with in-situ data analysis and visualization, to enable users to steer their next iteration of simulation or data collection.
The door to real-time HPC is starting to be unlocked – SGI, Intel, and CTC will continue to push that door open
I’d like to acknowledge all the participants in this work:
• Stephen Hawking CTC team
– James Briggs, Juha Jaykka, Paul Shellard, Carlos Martins (University of Porto)
• SGI team
– Karl Feind, Mike Woodacre, Simon Appleby
• Intel team
– John Pennycook, Johannes Guenther, Jefferson Amstutz, Ingo Wald, Jim Jeffers