Research Research Projects

SIRIUS: Next generation multitier data storage and I/O systems

Computational simulation has changed the way that we do science, but there is already a wide gulf between the speed with which we can compute and the rate and capacity with which we can store data. Data storage and management will be the limiting factor for exascale computers and beyond. SIRIUS is a multidisciplinary project conducting research into how this problem can be addressed by best utilising the many different types of storage media that are available on a typical high-performance computer, ranging from high speed in situ memory through to tape. The need to effectively use the entire multi-tier storage hierarchy requires the development and mathematical analysis of new mathematical techniques for data compression and refactoring techniques through to challenges in developing new software infrastructures that enable the computational scientist to utilise the multi-tier hierarchy in a seamless fashion.  

Research Leads: 

Scott Klasky (Oak Ridge National Laboratory)
Mark Ainsworth (Brown)
Jay Lofstead (Sandia)
Carlos Maltzahn (UC Santa Cruz)
Manish Parashar (Rutgers)

Funding Sources: 

Department of Energy
The figure shows the results of using the popular decimation technique to compress the primary variables φ and B in a fusion simulation. Whilst the main features appear to be preserved, if one looks at the derived variables, in this case the fluid velocity, the features are lost when one uses a uniform decimation scheme. This highlights the imperative to understand the implications of reduced data representations on down-stream analysis and visualization operations, and the unanticipated implications of reduced on data.