Large Web search engines are complex systems that solve thousands of user queries per second on clusters of dedicated distributed memory processors. Processing each query involves executing a number of operations to get the answer presented to the user. The most expensive operation in running time is the calculation of the top-k documents that best match each query. In this paper we propose the parallelization of a state of the art document ranking algorithm called Block-Max WAND. We propose a 2-steps parallelization of the WAND algorithm in order to reduce inter-processor communication and running time cost. Multi-threading tailored to Block-Max WAND is also proposed to exploit multi-core parallelism in each processor. The experimental results show that the proposed parallelization reduces execution time significantly as compared against current approaches used in search engines. © 2013 Springer-Verlag.
CITATION STYLE
Rojas, O., Gil-Costa, V., & Marin, M. (2013). Efficient parallel block-max WAND algorithm. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 8097 LNCS, pp. 394–405). https://doi.org/10.1007/978-3-642-40047-6_41
Mendeley helps you to discover research relevant for your work.