An enhanced dynamic replica creation and eviction mechanism in data grid federation environment

• Main Author: Argungu, Musa Sule
• Format: Thesis
• Language: English; English; English
• Published: 2018
• Subjects: QA71-90 Instruments and machines
• Online Access: https://etd.uum.edu.my/7461/1/Depositpermission_s900066.pdf; https://etd.uum.edu.my/7461/2/s900066_01.pdf; https://etd.uum.edu.my/7461/3/s900066_02.pdf; https://etd.uum.edu.my/7461/; http://sierra.uum.edu.my/record=b1697850~S1
• Institution: Universiti Utara Malaysia

Data Grid Federation system is an infrastructure that connects several grid systems, which facilitates sharing of large amount of data, as well as storage and computing resources. The existing mechanisms on data replication focus on finding file values based on the number of files access in deciding which file to replicate, and place new replicas on locations that provide minimum read cost. DRCEM finds file values based on logical dependencies in deciding which file to replicate, and allocates new replicas on locations that provide minimum replica placement cost. This thesis presents an enhanced data replication strategy known as Dynamic Replica Creation and Eviction Mechanism (DRCEM) that utilizes the usage of data grid resources, by allocating appropriate replica sites around the federation. The proposed mechanism uses three schemes: 1) Dynamic Replica Evaluation and Creation Scheme, 2) Replica Placement Scheme, and 3) Dynamic Replica Eviction Scheme. DRCEM was evaluated using OptorSim network simulator based on four performance metrics: 1) Jobs Completion Times, 2) Effective Network Usage, 3) Storage Element Usage, and 4) Computing Element Usage. DRCEM outperforms ELALW and DRCM mechanisms by 30% and 26%, in terms of Jobs Completion Times. In addition, DRCEM consumes less storage compared to ELALW and DRCM by 42% and 40%. However, DRCEM shows lower performance compared to existing mechanisms regarding Computing Element Usage, due to additional computations of files logical dependencies. Results revealed better jobs completion times with lower resource consumption than existing approaches. This research produces three replication schemes embodied in one mechanism that enhances the performance of Data Grid Federation environment. This has contributed to the enhancement of the existing mechanism, which is capable of deciding to either create or evict more than one file during a particular time. Furthermore, files logical dependencies were integrated into the replica creation scheme to evaluate data files more accurately.