FiberSCIP—A Shared Memory Parallelization of SCIP

Abstract

Recently, parallel computing environments have become significantly popular. In order to obtain the benefit of using parallel computing environments, we have to deploy our programs for these effectively. This paper focuses on a parallelization of SCIP (Solving Constraint Integer Programs), which is a MIP solver and constraint integer programming framework available in source code. There is a parallel extension of SCIP named ParaSCIP, which parallelizes SCIP on massively parallel distributed memory computing environments. This paper describes FiberSCIP, which is yet another parallel extension of SCIP to utilize multi-threaded parallel computation on shared memory computing environments, and has the following contributions: First, the basic concept of having two parallel extensions and the relationship between them and the parallelization framework provided by UG (Ubiquity Generator) is presented, including an implementation of deterministic parallelization. Second, the difficulties to achieve a good performance that utilizes all resources on an actual computing environment and the difficulties of performance evaluation of the parallel solvers are discussed. Third, a way to evaluate the performance of new algorithms and parameter settings of the parallel extensions is presented. Finally, current performance of FiberSCIP for solving mixed-integer linear programs (MIPs) and mixed-integer non-linear programs (MINLPs) in parallel is demonstrated.

Model Selection by Balanced Identification: the Interplay of Optimization and Distributed Computing

“…It is possible to use the SCIP solver implementing a branch-and-bound algorithm for mixedinteger nonlinear mathematical programming problems, but at the cost of increased solving time (and more memory consumption). Although parallel extensions of SCIP are available [30,31] simple workarounds are used now. They are: Ipopt multi starts options; some easily verifiable necessary conditions of the global optimum.…”

The SCIP Optimization Suite 8.0

“…UG has been developed over 10 years as beta versions to have general interfaces for the base solvers. Internally, we have developed parallel solvers for SCIP [100,103,101], CPLEX (not developed anymore), FICO Xpress [102], PIPS-SBB [77,78], Concorde 9 , and QapNB [29]. In addition to the parallelization of these branch-and-bound base solvers, UG was used to develop MAP-SVP [105], which is a solver for the Shortest Vector Problem (SVP), and whose algorithm does not rely on branch-and-bound.…”

“…The ramp-up is a process which runs until all CPU cores become busy. For a general discussion of the ramp-up process for parallel branch-and-bound, see Ralphs et al [86] and for the ramp-up process of FiberSCIP see Shinano et al [103]. One of the distinguishing features of FiberSCIP is racing ramp-up.…”

Assessing the Effectiveness of (Parallel) Branch-and-bound Algorithms

“…In both cases, the modular implementation enables the use of the same underlying sequential implementation with different parallel frameworks or the same parallel framework with different underlying sequential implementations. The UG framework, for example, has been customized to work with SCIP [Gamrath et al, 2016, 2018b, Xpress [FICO, Shinano et al, 2018a], and PIPS-SBB [Munguia et al, 2016, Munguía et al, 2019.…”