Independent Study Report: Propose Partial Replication Schemes for Replicated Declustering and Compare Their Performance

In this course, under the supervision of our professor, we have focused on the techniques for declustering data into multiple disks. Our aim is to get familiar with all the related work done till today so we have started from the very beginning of declustering which is done on relational databases and Cartesian product files [4, 5]. We have explored the literature and find out the most outstanding, milestone quality papers about declustering, read them and have small discussions each weak on one or two papers. This way we tried to understand the increased interest and need on declustering of data to multiple disks, which was parallel to the increase in the amount of data used in today’s high performance, high complexity computing environments. Those environments utilize multiple disks controlled with multiple processors compared to single disk and single processor machines used at the early times of declustering research. Today, algorithms and approaches have adjusted to multiple disk systems that generally store spatial data and frequently receive range queries. Today’s database systems manipulate terabytes of data related to different fields of sciences like cartography, epidemiology, transportation and Geographical Information Systems [10]. We have observed several approaches aimed at decreasing the amount of time required for retrieval of data under different types of queries. Most of the queries fall into one of the groups; range query, arbitrary query and connected query. Those queries are investigated and several algorithms proposed for each of them separately in the literature. Our initial reference point for partial replication schemes was the research paper of our professor which analyses and compares all of the replicated declustering schemes proposed till now. We examined the comparisons of different methods for replicated declustering under different query types and loads.