通过构建学术云计算基础设施挽回规划不周的HPC系统数百万美元投资的尝试:一个关于错误与迟来学习的故事
Attempt to Salvage Multi-million Dollars of Ill-conceived HPC System Investment by Creating Academic Cloud Computing Infrastructure. A Tale of Errors and Belated Learning
AI总结:
本研究针对华沙大学ICM中心OCEAN项目中部分HPC系统规划不佳、投资回报率低的问题,将两套系统重构为混合多云学术云A-CHOICeM,大幅拓展用户群体,同时梳理了设计流程与实施障碍。
摘要:
2015年,华沙大学跨学科数学与计算建模中心(ICM)作为耗资1.68亿兹罗提(3600万欧元)的OCEAN项目的一部分,建成了一座现代化数据中心并部署了三套大型HPC系统。其中部分系统规划不周、架构设计糟糕,在五年生命周期内带来的投资回报率极低。本文报告了一项为期两年的密集工作:将其中两套HPC系统重构为名为A-CHOICeM(Akademicka CHmura Obliczeniowa ICM,ICM学术计算云)的混合多云解决方案。该工作旨在将ICM典型HPC系统的用户群从约200至500人,拓展至华沙地区所有高等教育机构的约10万名潜在普通学术用户。该方案的主要特点包括:将本地部署的ICM云与多家公有云服务商集成、为特定学术用户群体构建定制化方案、容器化,以及集成AI和量子计算等特殊计算范式。本文介绍了方案设计的完整流程、与供应商的竞争性磋商过程,以及方案的最终完整规格。文中还阐述了过程中遇到的若干障碍、陷阱与困难,包括“老派”HPC管理员的保守态度、大学的官僚作风、国家资助政策及其他问题。
英文摘要:
In 2015 the Interdisciplinary Centre for Mathematical and Computational Modelling (ICM), University of Warsaw built a modern datacenter and installed three substantial HPC systems as part of a 168 M PLN (36 M Euro) OCEAN project. Some of the systems were ill-conceived, badly architected and for the five years of their life span have brought minimal ROI. This paper reports on a two-year intensive effort to reengineer two of these HPC systems into a hybrid, multi-cloud solution called A-CHOICeM (Akademicka CHmura Obliczeniowa ICM). The intention was to expand the user base of ICM typical HPC system from around 200 to 500 to about 100,000 potential general academic users from all institutes of higher learning in the Warsaw area. The main characteristics of this solution are integration of on-premises ICM Cloud with several public cloud providers, building solution tailored to particular groups of academic users, containerization, integration of special computational paradigms like AI and Quantum Computing. Full process of designing the solution, competitive dialogue with suppliers, and full final specifications for the solution are presented. Several roadblocks, pitfalls and difficulties encountered along the way, including the conservative attitude of "the old school" HPC admins, University bureaucracy, national funding policies and others are presented.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。