TY - GEN
T1 - Optimizing multiple queries on scientific datasets with partial replicas
AU - Weng, Li
AU - Catalyurek, Umit
AU - Kurct, Tahsin
AU - Agrawal, Gagan
AU - Saltz, Joel
PY - 2007
Y1 - 2007
N2 - We propose strategies to efficiently execute a query work-load, which consists of multiple related queries submitted against a scientific dataset, on a distributed-memory system in the presence of partial dataset replicas. Partial replication re-organizes and re-distributes one or more subsets of a dataset across the storage system to reduce I/O overheads and increase I/O parallelism. Our work targets a class of queries, called range queries, in which the query predicate specifies lower and upper bounds on the values of all or a subset of attributes of a dataset. Data elements whose attribute values fall into the specified bounds are retrieved from the dataset. If we think of the attributes of a dataset forming multi-dimensional space, where each attribute corresponds to one of the dimensions, a range query defines a bounding box in this multidimensional space. We evaluate our strategies in two scenarios involving range queries. The first scenario represents the case in which queries have overlapping regions of interest, such as those arising from an exploratory analysis of the dataset by multiple users. In the second scenario, queries represent adjacent rectilinear sections that capture an irregular subregion in the multi-dimensional space. This scenario corresponds to a case where the user wants to query and retrieve a spatial feature from the dataset. We propose cost models and an algorithm for optimizing such queries. Our results using queries for subsetting and analysis of medical image datasets show that effective use of partial replicas can result in reduction in query execution times.
AB - We propose strategies to efficiently execute a query work-load, which consists of multiple related queries submitted against a scientific dataset, on a distributed-memory system in the presence of partial dataset replicas. Partial replication re-organizes and re-distributes one or more subsets of a dataset across the storage system to reduce I/O overheads and increase I/O parallelism. Our work targets a class of queries, called range queries, in which the query predicate specifies lower and upper bounds on the values of all or a subset of attributes of a dataset. Data elements whose attribute values fall into the specified bounds are retrieved from the dataset. If we think of the attributes of a dataset forming multi-dimensional space, where each attribute corresponds to one of the dimensions, a range query defines a bounding box in this multidimensional space. We evaluate our strategies in two scenarios involving range queries. The first scenario represents the case in which queries have overlapping regions of interest, such as those arising from an exploratory analysis of the dataset by multiple users. In the second scenario, queries represent adjacent rectilinear sections that capture an irregular subregion in the multi-dimensional space. This scenario corresponds to a case where the user wants to query and retrieve a spatial feature from the dataset. We propose cost models and an algorithm for optimizing such queries. Our results using queries for subsetting and analysis of medical image datasets show that effective use of partial replicas can result in reduction in query execution times.
UR - http://www.scopus.com/inward/record.url?scp=47249120809&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=47249120809&partnerID=8YFLogxK
U2 - 10.1109/GRID.2007.4354141
DO - 10.1109/GRID.2007.4354141
M3 - Conference contribution
AN - SCOPUS:47249120809
SN - 1424415608
SN - 9781424415601
T3 - Proceedings - IEEE/ACM International Workshop on Grid Computing
SP - 259
EP - 266
BT - Proceedings - 8th IEEE/ACM International Conference on Grid Computing, GRID 2007
T2 - 8th IEEE/ACM International Conference on Grid Computing, GRID 2007
Y2 - 19 September 2007 through 21 September 2007
ER -