Energy minimization for real-time systems with non-convex and discrete operation modes
Dabiri, F.
Vahdatpour, A.
Potkonjak, M.
Sarrafzadeh, M.
Comput. Sci. Dept., Univ. of California, Los Angeles, CA;
Abstract
We present an optimal methodology for dynamic voltage scheduling problem in the presence of realistic assumption such as leakage-power and intra-task overheads. Our contribution is an optimal algorithm for energy minimization that concurrently assumes the presence of (1) non-convex energy-speed models as opposed to previously studied convex models, (2) discrete set of operational modes (voltages) and (3) intra-task energy and delay overhead. We tested our algorithm on MediaBench and task sets used in previous papers. Our simulation results show an average of 22% improvement in energy reduction in comparison with optimal algorithms for convex models without switching overhead and on average of 24% with consideration for energy and delay overheads. This analysis lays the groundwork for improving functionality in CAD design through non-convex techniques for discrete models.
Index
Terms
Available to subscribers and IEEE members.
References
Available to subscribers and IEEE members.
Citing Documents
Available to subscribers and IEEE members.
Cart
