Content area
Abstract
A parallel database system is a DBMS which is implemented on a tightly-coupled multiprocessor machine and exploits data partitioning. This thesis tries to compare system performances of parallel database with different configurations of concurrency controls and replication strategies. We also conducted experimental analysis of our new load balancing algorithm. Our parallel database simulation model is based on shared-nothing architecture.
In our parallel database model, data can be uniformly distributed or partially replicated among existing nodes. During running, the system will periodically check workload in each node. It will run data reallocation module to redistribute the existing local databases by moving one data block from one node to another on the fly. When a block is being moved from one processor node to another, the system only keeps other transactions from accessing that block. All other parts of the database are available. We believe this algorithm could be a candidate for solving load balancing problem in a parallel database system.