作者: Grégory Mounié , Pierre Neyron , Guillaume Huard , Cyrille Martin , Nicolas Capit
DOI:
关键词: Computer science 、 Priority scheduling 、 Robustness (computer science) 、 Grid 、 Scalability 、 Distributed computing 、 Programming complexity 、 Queue
摘要: In this article we present the design choices and evaluation of a batch scheduler for large clusters, named OAR. This is based upon an original that emphasizes on low software complexity by using high level tools. The global architecture built scripting language Perl relational database engine Mysql. goal project OAR to prove it possible today build complex system ressource management such tools without sacrificing efficiency scalability. Currently, our offers most important features implemented other schedulers as priority scheduling (by queues), reservations, backfilling some computing support. Despite use tools, experiments show has performances close systems. Furthermore, currently exploited 700 nodes (a metropolitan GRID) shown good robustness.