Run-time Resource Management in CMPs Handling Multiple Aging Mechanisms

Run-time resource management is fundamental for efficient execution of workloads on Chip Multiprocessors. Application- and system-level requirements (e.g. performance vs. power lifetime reliability) are generally conflicting each other, any decision assignment, such as core allocation or frequency tuning, may positively affect some them while penalizing others. Resource assignment decisions can be perceived in few instants time consumption, but not reliability. In fact, this latter changes very slowly based the accumulation effects various over a long horizon. Moreover, aging mechanisms have different causes; most them, Electromigration (EM), subject to temperature levels, Thermal Cycling (TC) caused mainly by variations (both amplitude frequency). Mitigating only EM negatively TC vice versa. We propose orchestration strategy balance consumption constraints short-term long-term. Experimental results show that proposed approach improves average Mean Time To Failure at least 17% 20% w.r.t. TC, respectively, providing same level nominal counterpart guaranteeing budget.