A New Scheduling Strategy for Solving the Motif Finding Problem on Heterogeneous Architectures

Abstract

The Motif Finding Problem is a well-known computationally intensive problem in bioinformatics. Exact solutions to finding a motif such as brute-force suffer from the intractability of their run time. To ameliorate this, different solutions relying on different hardware technologies have been proposed including the usage of FPGA or multicore architectures. Recently, deploying heterogeneous architectures that involve CPUs, GPUs, and FPGA kits seems to be very promising. The main goal is to speed up processing in order to achieve the result in a moderate time. The scheduling strategy dealing with such heterogeneity is one of the most important factors affecting the performance of the heterogeneous systems. In this paper, we introduce a new scheduling strategy that depends on the expected finish time when the problem is solved by only one type of architecture. The proposed scheduling strategy provides the fastest available processor type with a specific size of data items and instructs it to perform the parallel operations. Other slower architectures will get different sizes of data that can be processed exactly in the same time granted to the fastest architecture. This operation will be iterated until the finish of all the processing operations required to get the motif. Evaluation results of solving the motif finding problem using our proposed approach on a set of heterogeneous architectures versus the implementation on an individual architecture are compared. Results show that the new scheduling strategy yields much more advantageous results over the implementation on individual architectures. We believe that the proposed strategy is a step towards a well-defined framework for implementing run-time systems that support heterogeneous architectures.