In parallel computation domain, graph coloring is widely studied in its own and represents a reference problem for scheduling of parallel tasks. Unfortunately, common graph coloring strategies usually focus on minimizing the number of colors without any concern for the sizes of each color class, thus producing highly skewed color class distributions. However, to guarantee efficiency in parallel computations, but also in other application contexts, it is important to keep the color classes highly balanced in their sizes. In this paper we address this challenging issue for large scale graphs, proposing a fast parallel MCMC heuristic for sparse graphs that randomly generates good balanced colorings provided that a sufficient number of colors are made available. We show its effectiveness through some numerical simulations on random graphs.
|Name||LECTURE NOTES IN COMPUTER SCIENCE|
|Workshop||12th IAPR-TC15 Workshop on Graph-Based Representations in Pattern Recognition, GbRPR 2019|
|Period||19/6/19 → 21/6/19|
- Balanced graph coloring
- Parallel algorithms
- Markov Chain Monte Carlo method
- Greedy colorer