容器技术在高性能计算环境中的应用

Application of Container Technology in High Performance Computing Environment

【应用背景】高性能计算环境中间件SCE部署在多个前端服务器,科学计算应用软件需要在多个超算系统安装,软件与基础编译环境间的适配问题亟需解决。【目的】解决高性能计算环境中SCE中间件与前端服务器、应用软件与超级计算系统之间的兼容性问题,简化软件部署流程。【方法】本文利用Docker技术构建了中间件SCE的容器,利用Singularity技术实现了应用软件的容器化,并在高性能计算环境中多个结点进行了部署验证。【结果】SCE容器能够在环境中正常提供服务,应用软件容器与宿主机性能相当,其中LAMMPS容器运行时间与宿主机误差不超过2.63%。【结论】利用容器技术实现了中间件及应用软件的容器化封装,解决了软件与部署环境的适配问题,降低了软件部署的复杂度,提高了软件管理及运维效率。

[Background]High performance computing environment middleware SCE is deployed in multiple servers,and scientific computing software needs to be installed in multiple heterogeneous clusters.Therefore,the problem of software adaptation to the basic compiling environment needs to be solved.[Objective]The purpose of the work is to solve the installation problem of SCE an d scientific computing software at heterogeneous servers in high performance computing environ ment,and simplify the software deployment process.[Methods]For the purpose of this work,we build SCE container and application software containers with Docker and Singularity technology and test the containers in several nodes of the high performance computing environment.[Results]The SCE container can run stably in the environment.The scientific computing software containers have roughly equivalent on-host performance with a deviation less than 2.63%for the execution time of LAMMPS.[Conclusions]We build containers of SCE and scientific computing software,optimize the deployment of software,and increase the efficiency of operation and maintenance.