Online model swapping for architectural simulation

Online model swapping for architectural simulation
复制标题

建筑模拟的在线模型交换

DOI:
10.1145/3457388.3458670
复制
发表时间:
2021
期刊:
The 18th International Conference on Computing Frontiers
影响因子:
--
通讯作者:
Beard, Jonathan
Beard, Jonathan
中科院分区:
--
文献类型:
--
作者:
Lavin, Patrick;Young, Jeffrey;Vuduc, Richard;Beard, Jonathan

文献摘要

参考文献

被引文献

相似文献

随着系统和应用程序变得越来越复杂,详细的计算机体系结构模拟需要越来越多的时间。较长的仿真时间会导致设计迭代变慢,从而迫使架构师在想要快速迭代设计时使用更简单的模型,例如电子表格。然而,简单的模型并不容易使用,因为架构师必须依靠直觉来选择代表性模型,并且从简单模型到详细的硬件模拟的路径并不总是清晰的。在这项工作中,我们提出了一种通过在线监控模拟行为并自动用更简单的统计近似值替换详细模型来弥合简单和详细模拟之间差距的方法。我们通过在开源模拟器 SVE-Cachesim 中实现该方法来展示该方法的潜力,以交换内存层次结构中的一级数据缓存 (L1D)。这一概念证明表明,我们的技术可以训练简单的模型来匹配 L1D 中的真实程序行为,并且可以交换它们,而不会对下游模型的性能产生破坏性的副作用。我们的模型在整个周期计数中仅引入 8% 的错误,同时用于超过 90% 的模拟,并且使用的模型每次缓存访问所需的计算量减少了两到八倍。
As systems and applications grow more complex, detailed computer architecture simulation takes an ever increasing amount of time. Longer simulation times result in slower design iterations which then force architects to use simpler models, such as spreadsheets, when they want to iterate quickly on a design. Simple models are not easy to work with though, as architects must rely on intuition to choose representative models, and the path from the simple models to a detailed hardware simulation is not always clear.In this work, we present a method of bridging the gap between simple and detailed simulation by monitoring simulation behavior online and automatically swapping out detailed models with simpler statistical approximations. We demonstrate the potential of our methodology by implementing it in the open-source simulatorSVE-Cachesimto swap out the level one data cache (L1D) within a memory hierarchy. This proof of concept demonstrates that our technique can train simple models to match real program behavior in theL1Dand can swap them in without destructive side-effects for the performance of downstream models. Our models introduce only 8% error in the overall cycle count, while being used for over 90% of the simulation and using models that require two to eight times less computation per cache access.
DOI: 10.1145/3631882.3631890
发表时间: 2003
期刊: Proceedings of the International Symposium on Memory Systems
影响因子: --
作者:
M. Hind;V. T. Rajan;P. Sweeney
通讯作者: P. Sweeney
DOI: 10.1007/s11222-016-9649-y
发表时间: 2017-05-01
影响因子: 2.2
作者:
Piironen, Juho;Vehtari, Aki
通讯作者: Vehtari, Aki
DOI: 10.2307/2531585
发表时间: 1988-09-01
期刊: BIOMETRICS
影响因子: 1.9
作者:
GAUCH, HG
通讯作者: GAUCH, HG