Given the increasing complexity of multi-cores architectures, a wide range of architecture parameters must be tuned to find the best tradeoffs in terms of multiple metrics such as energy and delay. Multi-objective exploration of the huge design space of next generation multi-core architectures cannot be anymore based on intuition and past experience of the design architects. Automatic Design Space Exploration is necessary to systematically support the exploration and the quantitative comparison of design alternatives in terms of multiple competing objectives. This talk addresses these problems by presenting MULTICUBE Explorer, an open-source tool to support automatic and fast multi-objective optimization of system-on-chip architectures. MULTICUBE Explorer provides a set of sampling and optimization techniques for finding Pareto points. Design of Experiments techniques are used to identify the experimentation plan where the set of tunable design parameters can vary. Response Surface Modeling techniques are used to obtain a prediction of the system behavior based on the set of training data generated by DoE. Current multi-core architectures expose a set of dynamic parameters which can be tuned at run-time to get a specified Quality of Service (QoS). In this talk, we propose an exploration framework for supporting the management of system resources through application monitoring and re-configuration. First, the proposed framework operates at design-time to identify a set of promising operating points representing the best power/performance tradeoffs. Then, the operating points are used at run-time to support resource management policies.
Monday, February 06, 2012
Free and open to the public