A Framework for Fine-Grain Resource Management in Heterogeneous Parallel Architectures
Parallel architectures (a.k.a multicores) are now the de facto computing component in nearly all computerized systems ranging from mobile phones via desktop/laptops to data centers. Apart from being able to run parallel workloads, multiple processors (cores) on a multicore chip can offer a high application throughput by running multiple independent applications in parallel with a variety of performance requirements ranging from hard real-time, via interactive to background services with low performance requirements. Unfortunately, multiple applications that run on a multicore share architectural resources such as processor, memory, and interconnect structures. A resource management framework in which application performance requirements can be explicitly stated and continuously tracked to control resource allocation is currently lacking. The proposed research aims at exploring design principles for a novel resource-management framework for heterogeneous multicore systems in which 1) application performance requirements are explicitly stated 2) progress towards stated objectives is tracked 3) and resources needed to meet the objectives are controlled to reduce power efficiency. The overall goal is to significantly improve utilization and power efficiency of architectural resources in future heterogeneous multicore architectures.
Per Stenström (contact)
Full Professor at Chalmers, Computer Science and Engineering (Chalmers), Computer Engineering (Chalmers)
Professor at Software Technology (Chalmers)
University of Gothenburg
Swedish Research Council (VR)
Funding Chalmers participation during 2013–2016