MapReduce is a powerful distributed data processing model that is currently adopted in a wide range of domains to efficiently handle large volumes of data, i.e., cope with the big data surge. In this paper, we propose an approach to formal derivation of the MapReduce framework. Our approach relies on stepwise refinement in Event-B and, in particular, the event refinement structure approach – a diagrammatic notation facilitating formal development. Our approach allows us to derive the system architecture in a systematic and well-structured way. The main principle of MapReduce is to parallelise processing of data by first mapping them to multiple processing nodes and then merging the results. To facilitate this, we formally define interdependencies between the map and reduce stages of MapReduce. This formalisation allows us to propose an alternative architectural solution that weakens blocking between the stages and, as a result, achieves a higher degree of parallelisation of MapReduce computations.
If you cannot see the document below, the PDF document is most likely not freely accessible. In this case, please try to access the document via this link.
% BibTex
@inproceedings{PereverzevaBFLT14,
author = {Inna Pereverzeva and
Michael J. Butler and
Asieh Salehi Fathabadi and
Linas Laibinis and
Elena Troubitsyna},
editor = {Yamine A{\"{\i}}t Ameur and
Klaus{-}Dieter Schewe},
title = {Formal Derivation of Distributed MapReduce},
booktitle = {Abstract State Machines, Alloy, B, TLA, VDM, and {Z} - 4th International
Conference, {ABZ} 2014, Toulouse, France, June 2-6, 2014. Proceedings},
series = {Lecture Notes in Computer Science},
volume = {8477},
pages = {238--254},
publisher = {Springer},
year = {2014},
url = {https://doi.org/10.1007/978-3-662-43652-3\_21},
doi = {10.1007/978-3-662-43652-3\_21},
timestamp = {Tue, 14 May 2019 10:00:50 +0200},
biburl = {https://dblp.org/rec/conf/asm/PereverzevaBFLT14.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}