presented at event SPAA '16: 28th ACM Symposium on Parallelism in Algorithms and Architectures Conference