CAJ | 학술논문

许多科学实验和工程应用会产生TB或PB级的数据.这样庞大的数据体需要分布在世界不同地点的众多计算资源协同处理,高性能网格为使这些数据高效传送到地理上不同的位置提供了理想的手段.当用户频繁的访问大量远程数据时,动态副本可以有效地减少网格中的带宽消耗和访问延迟.在多层数据网格架构基础上提出了一种新型的动态副本策略:快速瀑布模型算法.OptorSim仿真器也被修改用以适应这种新型的策略.仿真结果表明,在随机访问模式下,对比Fast Spread,LRU和Economy-Zipf,快速瀑布策略能有效地减少平均数据访问时间,并且其优势随着用户提交作业数量的增加而增大.
허다과학실험화공정응용회산생TB혹PB급적수거.저양방대적수거체수요분포재세계불동지점적음다계산자원협동처리,고성능망격위사저사수거고효전송도지리상불동적위치제공료이상적수단.당용호빈번적방문대량원정수거시,동태부본가이유효지감소망격중적대관소모화방문연지.재다층수거망격가구기출상제출료일충신형적동태부본책략:쾌속폭포모형산법.OptorSim방진기야피수개용이괄응저충신형적책략.방진결과표명,재수궤방문모식하,대비Fast Spread,LRU화Economy-Zipf,쾌속폭포책략능유효지감소평균수거방문시간,병차기우세수착용호제교작업수량적증가이증대.
Scientific experiments and engineering applications generate large amounts of data which need to be shared among researchers around the world.High performance grids facilitate the distribution of such data to geographically remote places.Dynamic replication is a practical and efficient method which can be used to reduce bandwidth consumption and access latency in high performance data grids where users require remote access to large files.In this paper,a new dynamic replication algorithm,Fast Cascading,is proposed for the multi-tier Data Grid.Simulator OptorSim has been modified to illustrate the performance of the replication algorithm.The simulation results show that this algorithm can reduce the average response time of data access greatly for random access pattern compared to Fast Spread,LRU and Economy-Zipf dynamic replication strategy and the gain increases with the number of submitted jobs.