Solving The Low Performance Problem Caused By The Un Balanced Workloads Of Parallel Database Systems

Good Essays
In this article, the authors proposed to use a new technique – “resource bricolage” to solve the low performance problem caused by the un-balanced workloads in parallel database systems.
When a parallel database system is first constructed, the set of machines are made identical, therefore, the default data partitioning strategy for this parallel database is uniform data partitioning, and will ignore the differences among machines. In this case, all these identical machines will have the same workload, which will end up with similar performance efficiency. However, when time goes by, in this parallel database system, new machines that are different from the original ones will be added; old machines will be reconfigured/upgraded or replaced. These changes will result in a heterogeneous parallel database (the set of machines varies a lot from each other, such as having different disk, CPU, memory and network resources). When this happens, the default uniform data partitioning method will still allocate data evenly onto each machine. Due to the heterogeneous feature of each machine, the same amount of data will be processed with varying amounts of time: same workload may overload the slow machines and under-utilize the powerful machine. Based on the fact that the slowest machine will determine the processing time in a parallel database system, the situation, described above, will significantly harm the overall database performance and at the same time, waste
Get Access