Parallelism in object-oriented query processing

Abstract

This paper investigates parallel query processing issues in object-oriented databases. We first introduce the query model to be used. We then identify three areas where parallelism can be exploited in the query model. The perf o m n c e of each of the three areas is measured by mapping the query model to a parallel architecture and using analytic comparisons to the case when query processing is performed on a single processor. Our results show that there is always some speed-up in query processing time for parallel processing of a query over single processor processing. [0] INTRODUCTIONWe discussed query processing issues in object-oriented databases for different classes of queries in [KIM% KLM891. Our assumption was that query processing is performed on a single processor. In this paper, we focus attention to parallel query processing issues in object-oriented databases by exploiting parallelism during query processing and using a parallel architecture to process the query. We then compare the query processing time when using a parallel processor versus when using a single processor. Our motivation is mainly due to our emphasis to increase the performance of object-oriented query processing.It is not our intention to discuss in detail the parallel methods for object-oriented databases. We are merely interested in identifying the situations in the query model in which parallelism can be exploited during query evaluation and observe if performance can be enhanced through parallel techniques. Our analytical results show that there is some speedup in query evaluation when parallel techniques are used. However, the speedup is marginal at best. This may be due to the Amdahl's Law [AMDA67], which states that for a program with serial fraction s, the maximum parallel speedup obtainable is bounded by l/s. This has led to the assertation that the serial fraction will dominate execution time for any large parallel ensemble of processors, limiting the advantages of the parallel approach. We will show in the query model that serial processing is a major component of parallel query processing.[AGRA84] demonstrated that the main limitation to parallelism in database machines is the available I/O bandwidth provided by the underlying disk subsystem containing the database. Unless this problem is investigated and tackled, the use of a large number of processing elements does not seem justified. PIC851 showed that one approach to increase the potential YO bandwidth is to replace a single large disk with several smaller units. They also introduced a data model that would permit the exploitation of a significant number of parallel disk units. Similary, [KHOS88] states that there are two solutions for solving the VO bottleneck. One is to increase the U 0 bandwidth through U 0 parallelism and the other is to minimize the number of VOs through a clustering method.An object-oriented query model can exploit parallelism following the strategy of [BICSS]. The strategy is to increase U 0 bandwidth through VO parallelism using multiple disk...

Performance analysis of parallel object-oriented query processing algorithms

“…Similarly, the classes comprising the class hierarchy are processed in parallel contributing to the class-hierarchy parallelism. The ideas presented in [19] are interesting. However, the paper presents some preliminary analyses and considers only queries in which the objects of a single target class are fetched in response to a query.…”

“…However, the study does not investigate the parallel architectures and algorithms for the execution of OO queries. [19] investigates parallel processing issues in OO databases. It is observed that there is always a significant sequential query processing regardless of the exploitation of parallelism due to the navigational properties of query evaluation.…”

“…The parallel algorithms take advantage of the concept of pipelined data flows, used by other systems mentioned earlier, in increasing parallelism. The different types of parallelism presented in [19] are exploited in the processing. A user query is processed by multiple concurrent waves, thereby, improving the parallelism and eliminating the complexities in the processing of OO data.…”

“…Researchers have recently proposed several techniques for the parallel processing of the data defined by a semantic and/or OO data model [10,11,3,19]. The AGM [10,11] represents and processes the database as a network of interrelated entities modeled by the entity-relationship model.…”

Collection-intersect join algorithms for parallel object-oriented database systems

“…The emerging between parallel technology and OODB has shown promising results [3], [5], [7], [11]. However, most research done in this area concentrated on path expression queries with pointer chasing.…”