Date of Award
Ph.D. in Engineering Science
Computer and Information Science
Philip J. Rhodes
As the size of scientific datasets continues to grow, we will not be able to store enormous datasets on a single grid node, but must distribute them across many grid nodes. The implementation of partial or incomplete replicas, which represent only a subset of a larger dataset, has been an active topic of research. Partial Spatial Replicas extend this functionality to spatial data, allowing us to distribute a spatial dataset in pieces over several locations. We investigate solutions to the partial spatial replica selection problems. First, we describe and develop two designs for an Spatial Replica Location Service (SRLS), which must return the set of replicas that intersect with a query region. Integrating a relational database, a spatial data structure and grid computing software, we build a scalable solution that works well even for several million replicas. In our SRLS, we have improved performance by designing a R-tree structure in the backend database, and by aggregating several queries into one larger query, which reduces overhead. We also use the Morton Space-filling Curve during R-tree construction, which improves spatial locality. In addition, we describe R-tree Prefetching(RTP), which effectively utilizes the modern multi-processor architecture. Second, we present and implement a fast replica selection algorithm in which a set of partial replicas is chosen from a set of candidates so that retrieval performance is maximized. Using an R-tree based heuristic algorithm, we achieve O(n log n) complexity for this NP-complete problem. We describe a model for disk access performance that takes filesystem prefetching into account and is sufficiently accurate for spatial replica selection. Making a few simplifying assumptions, we present a fast replica selection algorithm for partial spatial replicas. The algorithm uses a greedy approach that attempts to maximize performance by choosing a collection of replica subsets that allow fast data retrieval by a client machine. Experiments show that the performance of the solution found by our algorithm is on average always at least 91% and 93.4% of the performance of the optimal solution in 4-node and 8-node tests respectively.
Tian, Yun, "Partial Replica Location And Selection For Spatial Datasets" (2013). Electronic Theses and Dissertations. 452.
Emphasis: Computer Science