CASS-MT Research Areas
Algorithms & Application Kernels
The Adaptive Supercomputing Initiative provides the best tools and techniques for application developers on multi-architecture hybrid HPC systems. The goal of this research is to uncover and resolve problems that arise from these environments. To do this it is necessary to port real and significant applications to these environments.
Tasks
- Advanced Social Network Analysis
- Statistical Textual Document Analysis
- Dynamic Network Analysis
- Sparse Graph Network-of-Network Algorithms
- Contingency Analysis
Advanced Systems Software
Applications that will run on the Cray XMT architecture will need to take advantage of its unique nature. The multithreaded ThreadStorm processors and the unique architecture of the XMT offer significant advantages for many application areas. However, the software environment of the Cray XMT is unique and often requires significant rethinking of how traditional software uses the underlying hardware. The main goal of the systems software research for CASS-MT is to target the particular needs of multithreaded applications to ensure they realize the performance benefits the XMT offers.
Tasks
High-Productivity Programming Models
The Cray XMT provides a unique, highly productive programming environment based on a sophisticated, parallelizing C/C++ compiler as well as associated performance and debugging tools. This activity focuses on extending the Cray XMT environment by adding new functionality to the existing programming model that can further enhance application productivity. We are also conducting research on how to address the particular features of hybrid computing systems (such as the XMT) in an integrated programming model.
Experiences with application performance on the Cray MTA-2 architecture will have to be refined for the XMT machine; its inter-node communication architecture is substantially different from the MTA-2. For instance, the XMT inter-node communication architecture is substantially different from the MTA-2; it builds upon the Cray XT-4’s interconnect, which is inherently designed for distributed memory systems and applications that exploit locality, with minimal adaptations for the global shared memory environment of the XMT. The XMT’s ThreadStorm processors are derived from the original MTA-2 processors with adaptations for supporting commodity memory subsystems.
In light of these architectural changes and the insights gained from our experimentation in the past we propose several enhancements to the MTA programming model and compilation process that should result in productivity and performance benefits. These extensions would increase Cray compiler effectiveness for certain classes of applications by better thread management, locality optimizations, and providing options for deterministic scheduling of the threads.
Tasks
Possible Future Applications
- Random or indirect memory accesses
- Dynamic or unbalanced subcomputations
- Unstructured, dynamic, and/or sparse data structures
- Linked data structures (lists, graphs, trees)
- Sorting or searching
- Applications that need to access large amounts of memory (terabytes) and in an unpredictable manner.
- Graph Analysis (intelligence, protein folding, bioinformatics)
- Optimization problems (branch-and-bound, linear programming)
- Computational geometry (graphics, scene recognition and tracking)
- Coupled physics with multiple materials and time scales
- Data mining
- Business intelligence
- Pattern matching
- Power grid analysis
- Rendering, etc.