论文标题
为将来的HPC系统轻松地部署容器化应用程序
Enabling EASEY deployment of containerized applications for future HPC systems
论文作者
论文摘要
即将到来的Exascale时代将推动从基于CPU的系统中基于CPU的系统中的混合GPU繁重系统中的计算架构变化，具有更高的复杂性水平。尽管预计此类簇将提高某些优化的HPC应用程序的性能，但对于那些尚未适应其代码或使用新的编程范式从头开始的用户，它也会增加困难。由于仍然没有全面的自动辅助机制来提高此类系统上的应用程序性能，因此我们为将来的HPC体系结构（称为oesey）提供了支持框架（为每个人启用ExasClae）。该解决方案构建在分层软件体系结构上，该体系结构为不同的调整任务提供了不同的机制。这使用户可以调整每个层上的参数，从而增强其代码的特定特征。我们使用基于CharlieCloud的解决方案介绍了该框架，并在我们框架的上层展示了Lulesh基准测试。我们的方法可以自动用可忽略不计的开销来自动部署优化的容器计算，同时减少科学家在手动作业提交配置上花费的时间。
The upcoming exascale era will push the changes in computing architecture from classical CPU-based systems in hybrid GPU-heavy systems with much higher levels of complexity. While such clusters are expected to improve the performance of certain optimized HPC applications, it will also increase the difficulties for those users who have yet to adapt their codes or are starting from scratch with new programming paradigms. Since there are still no comprehensive automatic assistance mechanisms to enhance application performance on such systems, we are proposing a support framework for future HPC architectures, called EASEY (Enable exASclae for EverYone). The solution builds on a layered software architecture, which offers different mechanisms on each layer for different tasks of tuning. This enables users to adjust the parameters on each of the layers, thereby enhancing specific characteristics of their codes. We introduce the framework with a Charliecloud-based solution, showcasing the LULESH benchmark on the upper layers of our framework. Our approach can automatically deploy optimized container computations with negligible overhead and at the same time reduce the time a scientist needs to spent on manual job submission configurations.
