Runtime Program Optimization and Adaptation on Emerging Multi-core Systems

This project investigates runtime program optimization and adaptation methods for problems pertaining to scheduling, data allocation and data layout, on emerging multi-processors built from multi-core processors and/or multi-threaded cores. The project develops support for a unified, explicit parallel programming model, with processor-independent, scale-independent and topology-independent abstractions of parallel execution. This effort has been supported through an IBM Faculty Award, an NSF CAREER award, an NSF ITR award, and an NSF MRI award. Details can be found here.

MELISSES - Continuous Monitors and Predictors for Dynamic Adaptation on Many-Core Systems

This project explores empirical models of dynamic performance, power and scalability prediction on multi-core and many-core parallel architectures. The project develops prediction methodologies that combine continuous feedback from real-time hardware performance monitors with machine learning methods. This effort is supported by a DOE Early Career Principal Investigator Award. Details are available here.

VT-ASOS

VT-ASOS is a new hypervisor framework for application-specific operating system customization and adaptation on multi-core and many-core HPC systems. See [STMCS07], [NSFNGS08], for a position paper and preliminary results.

Thermal Conductors

Thermal conductors is a software framework for runtime thermal characterization and optimization of parallel programs on multi-core systems. Details coming soon.