Parallel Graph Algorithms Through Approximation

Md Ariful Hasan Khan, Purdue University

Abstract

We consider the problem of computing a b-MATCHING and a b-EDGE COVER, which are subgraphs of a graph with specific properties. Although there are polynomial-time algorithms for these problems, exact algorithms are impractical for massive graphs with billions or more of edges. Hence we design approximation algorithms that have fast run time complexity on serial computers. Since computations on massive graphs are compute-intensive, we also design approximation algorithms that possess a high degree of concurrency, so that they can be implemented efficiently on shared-and distributed-memory multiprocessors. The b-MATCHING problem is a generalization of the well-known Matching problem in graphs, where the objective is to choose a subset of M edges in the graph such that at most a specified number b(v) of edges in M are incident on each vertex v. Subject to this restriction, we maximize the sum of the weights of the edges in M. We describe a half-approximation algorithm, b-SUITOR, for computing a b-MATCHING of maximum weight in a graph with weights on the edges. Our results show that the b-SUITOR algorithm outperforms previously known algorithms, the GREEDY and LD (locally dominant edge) algorithms, by one to two orders of magnitude on a serial processor. The b-SUITOR algorithm has a high degree of concurrency, and it scales well up to 240 threads on a shared memory multiprocessor. We also implement the algorithm in distributed memory settings using a hybrid strategy where inter-node communication uses MPI and intra-node computation is done with OpenMP threads. We demonstrate strong and weak scaling of b-SUITOR up to 16K processors on two supercomputers at NERSC. We compute a b-MATCHING in a graph with 2 billion edges in under 4 seconds using 16K processors. The b-EDGE COVER problem is a generalization of the better-known EDGE COVER problem in graphs, where the objective is to choose a subset C of edges in the graph such that at least a specified number b(v) of edges in C are incident on each vertex v. In the weighted b -EDGE COVER problem, we minimize the sum of the weights of the edges in C. We design three new approximation algorithms for the b-EDGE COVER problem, and compare them with the previously known GREEDY algorithm. At each step, the GREEDY algorithm updates the effective weights of the edges, and adds an edge of minimum effective weight to the current edge cover. The updates of the effective weights makes the GREEDY algorithm sequential and impractical for massive graphs. A second algorithm, the LSE (locally sub-dominant edge) algorithm, adds edges with minimum effective weight in its neighborhood to the current cover, and it is amenable for parallelization. The LSE algorithm computes the same edge cover as the GREEDY algorithm, and both are 3/2-approximation algorithms. We design a third algorithm, S-LSE, an extension of the LSE algorithm, which uses static edge weights instead of dynamic effective weights used by the latter. This relaxation causes S-LSE to have a worse approximation ratio of 2, but makes it more amenable for efficient implementation and parallelization. A fourth algorithm, the MCE (matching complement edge cover) algorithm, is obtained from a relationship between approxi- mation algorithms for the b-EDGE COVER and the b-MATCHING problems. We prove that both S-LSE and MCE algorithms compute the same b-EDGE COVER, and hence both have approximation ratios of 2. In practice, all these algorithms compute edge covers with weights that are close to the optimal b-EDGE COVER, and have weights within 10% of each other. We parallelize and report results from the three new approximation algorithms, LSE, S-LSE and MCE in the context of shared memory multicore machines. Our results show that the MCE algorithm is faster than the other algorithms by at least an order of magnitude, both on serial and shared memory multiprocessors.

Degree

Ph.D.

Advisors

Pothen, Purdue University.

Subject Area

Computer science

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS