Adaptive Massively Parallel Algorithms for Cut Problems

TL;DR

This paper introduces a groundbreaking sublogarithmic AMPC algorithm for approximate weighted Min Cut, leveraging adaptive access to surpass previous MPC limitations and achieve faster convergence.

Contribution

It presents the first $O( ext{log log } n)$ round AMPC algorithm for weighted Min Cut, breaking the $O( ext{log } n)$ barrier under the 1-vs-2 Cycle Conjecture.

Findings

Achieves $O( ext{log log } n)$ rounds for approximate Min Cut in AMPC.

Decouples divide and conquer layers for exponential speedup.

Fully scalable with local memory $O(n^ ext{epsilon})$.

Abstract

We study the Weighted Min Cut problem in the Adaptive Massively Parallel Computation (AMPC) model. In 2019, Behnezhad et al. [3] introduced the AMPC model as an extension of the Massively Parallel Computation (MPC) model. In the past decade, research on highly scalable algorithms has had significant impact on many massive systems. The MPC model, introduced in 2010 by Karloff et al. [16], which is an abstraction of famous practical frameworks such as MapReduce, Hadoop, Flume, and Spark, has been at the forefront of this research. While great strides have been taken to create highly efficient MPC algorithms for a range of problems, recent progress has been limited by the 1-vs-2 Cycle Conjecture [20], which postulates that the simple problem of distinguishing between one and two cycles requires $\Omega(\log n)$ MPC rounds. In the AMPC model, each machine has adaptive read access to a distributed hash table even when communication is restricted (i.e., in the middle of a round). While remaining practical [4], this gives algorithms the power to bypass limitations like the 1-vs-2 Cycle Conjecture. We give the first sublogarithmic AMPC algorithm, requiring $O(\log\log n)$ rounds, for $(2+\epsilon)$-approximate weighted Min Cut. Our algorithm is inspired by the divide and conquer approach of Ghaffari and Nowicki [11], which solves the $(2+\epsilon)$-approximate weighted Min Cut problem in $O(\log n\log\log n)$ rounds of MPC using the classic result of Karger and Stein [15]. Our work is fully-scalable in the sense that the local memory of each machine is $O(n^\epsilon)$ for any constant $0 < \epsilon < 1$. There are no $o(\log n)$-round MPC algorithms for Min Cut in this memory regime assuming the 1-vs-2 Cycle Conjecture holds. The exponential speedup in AMPC is the result of decoupling the different layers of the divide and conquer algorithm and solving all layers in $O(1)$ rounds.