Integration of ParaSCIP solvers running on several clusters on the base of Everest cloud platform

Abstract

Abstract Software integration of optimization problems’ solvers leveraging power of heterogeneous computing environments is a great challenge of last decades. Last several years we have been developing coarse-grained parallelization approaches to speed up Branch-and-Bound (BnB) algorithm for discrete and global optimization problems by exchange of BnB-incumbents in a heterogeneous environment containing standalone servers and clusters via Everest software toolkit, http://everest.distcomp.org. This approach have been implemented as DDBNB Everest-application (Domain Decomposition BnB), https://github.com/distcomp/ddbnb. The current implementation is based on two solvers (and their open API to get/put incumbents): SCIP, https://scip.zib.de and CBC, https://github.com/coin-or/Cbc. Recently we began to use ParaSCIP solver, https://ug.zib.de, - parallel implementation of BnB-algorithm in SCIP based on MPI. ParaSCIP demonstrates an advantages of fine-grained parallelization in homogeneous computing environment, i.e. HPC-clusters. By now we have access to three clusters from Russian Top50 where ParaSCIP have been installed. In the article several ways to involve ParaSCIP processes running on different clusters in solving common optimization problem are discussed.