Macro-scale simulation has been advanced as one tool for application-architecture co-design to express operation of exascale systems. These simulations approximate the behavior of system components, trading off accura...
详细信息
ISBN:
(纸本)9781479989379
Macro-scale simulation has been advanced as one tool for application-architecture co-design to express operation of exascale systems. These simulations approximate the behavior of system components, trading off accuracy for increased evaluation speed. Application skeletons serve as the vehicle for these simulations, but they require accurately capturing the execution behavior of computation. The complexity of application codes, the heterogeneity of the platforms, and the increasing importance of simulating multiple performance metrics (e. g., execution time, energy) require new modeling techniques. We propose flexible statistical models to increase the fidelity of application simulation at scale. We present performance model validation for several exascale mini-applications that leverage a variety of parallel programming frameworks targeting heterogeneous architectures for both time and energy performance metrics. When paired with these statistical models, application skeletons were simulated on average 12.5 times faster than the original application incurring only 6.08% error, which is 12.5% faster and 33.7% more accurate than baseline models.
Successful HPC over desktop grids and non-dedicated NOWs is challenging, since good performance is difficult to achieve due to dynamic workloads. On iterative data-parallel applications, this is addressed by dynamic d...
详细信息
ISBN:
(纸本)9780769534718
Successful HPC over desktop grids and non-dedicated NOWs is challenging, since good performance is difficult to achieve due to dynamic workloads. On iterative data-parallel applications, this is addressed by dynamic data distribution. However, current approaches migrate an application from one distribution to another in one single phase, which can impact performance. In this paper, we present D-3-ARC, a programming framework to support adaptive and incremental data distribution, so that data migration takes place over several successive iterations. D-3-ARC consists of a runtime system and an API for specifying the distribution of arrays as well as how data redistribution takes place. We demonstrate how D-3-ARC can be used to develop an incremental strategy for data distribution in a Poisson solver, utilising a runtime feedback mechanism to determine how much data to migrate during each iteration.
暂无评论