CSE Dissertation Proposal Defense: James Orr

Jan 26, 2018
2 p.m.
4 p.m.
Jolley Hall, Room 309

"Scheduling and Platforms for Adaptive Parallel Real-Time Systems"

James Orr
Adviser: Chris Gill

Adaptive scheduling techniques such as the elastic task model have been used in sequential real-time systems to schedule dynamic and stochastic tasks. However, they have not yet been adapted to parallel real-time tasks or to concurrency platforms that support them. To support a new generation of adaptive parallel real-time systems, I propose to generalize and expand support for task elasticity in parallel real-time scheduling techniques, including Federated Scheduling and Semi-Federated Scheduling. I also propose to develop platform support for those techniques so that the rates at which parallel real-time tasks are released, and the numbers of cores on which they run, can be adapted dynamically at run-time.

To validate and leverage these new capabilities, I then propose to advance the state of the art in real-time hybrid simulation (RTHS), a currently static application of parallel real-time computing that is of particular interest in structural and earthquake engineering. Dynamic reallocation of resources allows for larger structures to be simulated under more realistic (intense and dynamic) workloads. Additionally I propose to explore how these new capabilities may improve mixed-criticality systems in which some tasks are innately more important than others, including for system recovery and for new system models in which the criticality of each task may change depending on the current mode of the system. I will realize these capabilities, and conduct these experiments by developing APaRTEC, a novel adaptive concurrency platform framework in which a wide variety of parallel real-time applications can adapt their computational and/or temporal resolution using a variety of parallel scheduling techniques.