Exploring the Performance of Singularity for High Performance Computing Scenarios
Exploring the Performance of Singularity for High Performance Computing Scenarios
复制标题
探索高性能计算场景的奇点性能
DOI：
--
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
Wenbo Chen
中科院分区：
文献类型：
--
作者：
Guangchao Hu;Yang Zhang;Wenbo Chen
传统的基于虚拟机管理程序的虚拟化解决方案由于性能开销而在高性能计算（HPC）中不常用。以Linux Container和Docker为代表的基于容器的虚拟化技术可以提供更好的资源共享、定制环境和低开销。然而，它们仍然不能满足HPC环境中的功能和安全需求。Singularity是一种有吸引力的基于容器的方法，可以满足科学应用的要求。它具有计算的移动性、可复制性、用户自由度和对现有传统HPC的支持等特点，可以有效解决其他虚拟化技术的一些缺陷。在本文中，我们进行了详细的性能评估，CPU，内存和网络带宽之间的Singularity和裸金属使用HPL，流和OSU微基准。通过部署NAMD、VASP、AMBER和WRF等四个典型HPC应用，进一步验证了Singularity的性能开销。此外，还研究了奇点的迁移和相容性。实验结果表明，Singularity在MPI和GPU并行应用中均能达到接近原生的性能，在高性能计算中具有较大的应用前景。
Traditional hypervisor-based virtualization solutions have not been commonly used in High Performance Computing (HPC) due to the performance overhead. Container-based virtualization technologies represented by Linux Container and Docker can provide better resource sharing, customized environments and low overhead. However, they still can't satisfy the functional and security needs in HPC environments. Singularity is an attractive container-based approach to meet the requirements of scientific applications. Its characteristics such as mobility of compute, reproducibility, user freedom and support on existing traditional HPC, can effectively solve some flaws compared to other virtualization technologies. In this paper, we conducted a detailed performance evaluation of CPU, memory and network bandwidth between Singularity and bare metal by using HPL, STREAM and OSU Micro benchmarks. Four typical HPC applications such as NAMD, VASP, AMBER, and WRF are deployed to further validate the performance overhead of Singularity. Furthermore, migration and compatibility of Singularity are also investigated. Results from the experiments show that Singularity can achieve near-native performance in MPI and GPU parallel applications, which also demonstrate that it has a greater application prospect in HPC.