Geographically-distributedmulticluster computer system: architecture and software
The geographically-distributed multicluster computer system (CS) has been created and isbeing developed by Computer centre for parallel technologies of Siberian state university for telecommunicationsand information sciences (CPCT SibSUTIS) in conjunction with Laboratory ofcomputer systems of A.V.Rzhanov Institute of semiconductor physics of Siberian branch of RussianAcademy of Sciences (ISP SBRAS). The system composes geographically-distributed clustersand system software. Architecture of the given system is based on results of Russian scientificschool on the distributed computer systems.1. Architecture of the geographically-distributed multicluster computer system. The currentsystem configuration unites 9 geographically-distributed cluster computer systems, seven ofwhich are located in CPCT SibSUTIS (geographically it is the center of Novosibirsk) and theother two in ISP SBRAS (Academgorodok). The system includes a cluster, which is a part of distributedinfrastructure of the program University cluster.2. Software. It includes: standard components and original toolkit for parallel multiprogramming.Standard software components include: network operating system (GNU/Linux), toolsfor developing, debugging and analyzing of serial and parallel programs (MPI library: MPICH2,OpenMPI), the software for organization of system functioning (batch processing systemTORQUE, scheduler MAUI), for interactions of the geographically-distributed clusters (GlobusToolkit) and for job dispatching (GridWay).The original toolkit includes: means for self-checking and self-diagnostics, a subsystem fororganization of system functioning in multiprogramming modes, including: the environment fornesting of parallel programs and implementation of effective collective communications betweentheir branches, the distributed job queue and the manager of user queries, tools for analysis ofparallel programs, services for monitoring and organization of remote access to the system resources.The system is used for doing researches in the area of distributed processing of the information,debugging of tools for parallel multiprogramming and preparation of experts in the area offault-tolerant computing technologies. The system has confirmed efficiency of architectural decisionsand perspective of formation of the regional computer systems and their application for thesolving of superchallenging jobs and for modeling of the modern technological processes and thenatural phenomena.
Keywords
распределенные вычислительные системы, GRID, параллельное мультипрограммирование, эффективное выполнение параллельных программ, system software, geographically-distributed computer systemsAuthors
Name | Organization | |
Khoroshevsky Viktor G. | Siberian state universityfor telecommunication and information sciences | khor@sibsutis.ru |
Kurnosov Mikhail G. | Siberian state universityfor telecommunication and information sciences | mkurnosov@gmail.com |
Mamoilenko Sergey N. | Siberian state universityfor telecommunication and information sciences | sergey@cpct.sibsutis.ru |
References
