Data parallel acceleration of decision support queries using cell/BE and GPUs

Abstract

Decision Support System (DSS) workloads are known to be one of the most time-consuming database workloads that processes large data sets. Traditionally, DSS queries have been accelerated using large-scale multiprocessor. The topic addressed in this work is to analyze the benefits of using high-performance/low- cost processors such as the GPUs and the Cell/BE to accelerate DSS query execution. In order to overcome the programming effort of developing code for different architectures, in this work we explore the use of a platform, Rapidmind, which offers the possibility of executing the same program on both Cell/BE and GPUs. To achieve this goal we propose data-parallel versions of the original database scan and join algorithms. In our experimental results we compare the execution of three queries from the standard DSS benchmark TPC-H on two systems with two different GPU models, a system with the Cell/BE processor, and a system with dual quad-core Xeon processors. The results show that parallelism can be well exploited by the GPUs. The speedup values observed were up to 21× compared to a single processor system. Copyright 2009 ACM.