Resumen de Towards resource-aware computing for task-based runtimes and parallel architectures
Current large scale systems show increasing power demands, to the point that it has become a huge strain on facilities and budgets. The increasing restrictions in terms of power consumption of High Performance Computing (HPC) systems and data centers have forced hardware vendors to include power capping capabilities in their commodity processors. Power capping opens up new opportunities for applications to directly manage their power behavior at user level. However, constraining power consumption causes the individual sockets of a parallel system to deliver different performance levels under the same power cap, even when they are equally designed, which is an effect caused by manufacturing variability. Modern chips suffer from heterogeneous power consumption due to manufacturing issues, a problem known as manufacturing or process variability. As a result, systems that do not consider such variability caused by manufacturing issues lead to performance degradations and wasted power. In order to avoid such negative impact, users and system administrators must actively counteract any manufacturing variability.
In this thesis we show that parallel systems benefit from taking into account the consequences of manufacturing variability, in terms of both performance and energy efficiency. In order to evaluate our work we have also implemented our own task-based version of the PARSEC benchmark suite. This allows to test our methodology using state-of-the-art parallelization techniques and real world workloads. We present two approaches to mitigate manufacturing variability, by power redistribution at runtime level and by power- and variability-aware job scheduling at system-wide level. A parallel runtime system can be used to effectively deal with this new kind of performance heterogeneity by compensating the uneven effects of power capping. In the context of a NUMA node composed of several multi core sockets, our system is able to optimize the energy and concurrency levels assigned to each socket to maximize performance. Applied transparently within the parallel runtime system, it does not require any programmer interaction like changing the application source code or manually reconfiguring the parallel system. We compare our novel runtime analysis with an offline approach and demonstrate that it can achieve equal performance at a fraction of the cost. The next approach presented in this theis, we show that it is possible to predict the impact of this variability on specific applications by using variability-aware power prediction models. Based on these power models, we propose two job scheduling policies that consider the effects of manufacturing variability for each application and that ensures that power consumption stays under a system wide power budget. We evaluate our policies under different power budgets and traffic scenarios, consisting of both single- and multi-node parallel applications.
