Characterization of Vertex-centric Breadth First Search for Lattice Graphs

Dindokar, Ravikant and Simmhan, Yogesh (2017) Characterization of Vertex-centric Breadth First Search for Lattice Graphs. In: EEE 24th International Conference on High Performance Computing Workshops (HiPCW), DEC 18-21, 2017, Jaipur, INDIA, pp. 32-39.

PDF HIPCW_32_2017.pdf - Published Version Restricted to Registered users only Download (580kB) | Request a copy

Abstract

Google's Pregel vertex-centric programming model has led to a class of component-centric abstractions that are widely used for designing and executing distributed graph algorithms in a Bulk Synchronous Parallel (BSP) manner on commodity clusters. There has, however, been little effort to formally understand the behavior of such programming models to allow analytical comparison of graph algorithms with narrow bounds. Graphs and graph algorithms have asymmetry that make them challenging to model effectively. Lattice (or grid) graphs, commonly used in scientific computing, offer a simpler structure to tackle this novel problem. Here, we propose analytical models with tight bounds on the coordination, computation and communication costs for performing a vertex-centric Breadth First Search (BFS) on uniform lattice graphs. We validate these further with empirical results. Our results hold for lattice graphs of arbitrary dimension and uniform odd-valued length, but can be extended to non-uniform lengths as well. These results offer a formal understanding of the behavior of vertex-centric BFS for a lattice structure, and lays the foundation for further analysis and exploration of component-centric algorithms.

Item Type:	Conference Proceedings
Additional Information:	Copy right for the article belong to IEEE, 345 E 47TH ST, NEW YORK, NY 10017 USA
Department/Centre:	Division of Interdisciplinary Sciences > Supercomputer Education & Research Centre
Date Deposited:	27 Mar 2018 18:27
Last Modified:	27 Mar 2018 18:27
URI:	http://eprints.iisc.ac.in/id/eprint/59406

Actions (login required)

View Item