Yuji Shinano

ParaSCIP - a parallel extension of SCIP

Mixed integer programming (MIP)has become one of the most important techniques in Operations Research and Discrete Optimization. SCIP (Solving Constraint Integer Programs) is currently one of the fastest non-commercial MIP solvers. It is based on the branchandboundprocedure in which the problem is recursively split into smaller subproblems, thereby creating a so-called branching tree. We present ParaSCIP, an extension of SCIP, which realizes a parallelization on a distributed memory computing environment. ParaSCIP uses SCIP solvers as independently running processes to solve subproblems (nodes of the branching tree) locally. This makes the parallelization development independent of the SCIP development. Thus, ParaSCIP directly profits from any algorithmic progress in future versions of SCIP. Using a first implementation of ParaSCIP, we were able to solve two previously unsolved instances from MIPLIB2003, a standard test set library for MIP solvers. For these computations, we used up to 2048 cores of the HLRN II supercomputer.

Could we use a million cores to solve an integer program

Given the steady increase in cores per CPU, it is only a matter of time before supercomputers will have a million or more cores. In this article, we investigate the opportunities and challenges that will arise when trying to utilize this vast computing power to solve a single integer linear optimization problem. We also raise the question of whether best practices in sequential solution of ILPs will be effective in massively parallel environments.

FiberSCIP—A Shared Memory Parallelization of SCIP

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 mixed-integer linear programming 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, we present the basic concept of having two parallel extensions, and the relationship between them and the parallelization framework provided by UG (Ubiquity Generator), including an implementation of deterministic parallelization. Second, we discuss th...

SCIP-Jack—a solver for STP and variants with parallelization extensions

The Steiner tree problem in graphs is a classical problem that commonly arises in practical applications as one of many variants. While often a strong relationship between different Steiner tree problem variants can be observed, solution approaches employed so far have been prevalently problem-specific. In contrast, this paper introduces a general-purpose solver that can be used to solve both the classical Steiner tree problem and many of its variants without modification. This versatility is achieved by transforming various problem variants into a general form and solving them by using a state-of-the-art MIP-framework. The result is a high-performance solver that can be employed in massively parallel environments and is capable of solving previously unsolved instances.

Solving Open MIP Instances with ParaSCIP on Supercomputers Using up to 80,000 Cores

This paper describes how we solved 12 previously unsolved mixed-integer programming (MIP) instances from the MIPLIB benchmark sets. To achieve these results we used an enhanced version of ParaSCIP, setting a new record for the largest scale MIP computation: up to 80,000 cores in parallel on the Titan supercomputer. In this paper we describe the basic parallelization mechanism of ParaSCIP, improvements of the dynamic load balancing and novel techniques to exploit the power of parallelization for MIP solving. We give a detailed overview of computing times and statistics for solving open MIPLIB instances.

Solving Hard MIPLIB2003 Problems with ParaSCIP on Supercomputers: An Update

Contemporary supercomputers can easily provide years of CPU time per wall-clock hour. One challenge of todays software development is how to harness this vast computing power in order to solve really hard mixed-integer programming instances. In 2010, two out of six open MIPLIB2003 instances could be solved by ParaSCIP in more than ten consecutive runs, restarting from checkpointing files. The contribution of this paper is threefold: For the first time, we present computational results of single runs for those two instances. Secondly, we provide improved upper and lower bounds for all of the remaining four open MIPLIB2003 instances. Finally, we explain which new developments led to these results and discuss the current progress of ParaSCIP. Experiments were conducted on HLRN II, on HLRN III, and on the Titan supercomputer, using up to 35,200 cores.

Parallel Solvers for Mixed Integer Linear Optimization

In this chapter, we provide an overview of the current state of the art with respect to solution of mixed integer linear optimization problems (MILPs) in parallel. Sequential algorithms for solving MILPs have improved substantially in the last two decades and commercial MILP solvers are now considered effective off-the-shelf tools for optimization. Although concerted development of parallel MILP solvers has been underway since the 1990s, the impact of improvements in sequential solution algorithms has been much greater than that which came from the application of parallel computing technologies. As a result, parallelization efforts have met with only relatively modest success. In addition, improvements to the underlying sequential solution technologies have actually been somewhat detrimental with respect to the goal of creating scalable parallel algorithms. This has made efforts at parallelization an even greater challenge in recent years. With the pervasiveness of multi-core CPUs, current state-of-the-art MILP solvers have now all been parallelized and research on parallelization is once again gaining traction. We summarize the current state-of-the-art and describe how existing parallel MILP solvers can be classified according to various properties of the underlying algorithm.

A First Implementation of ParaXpress: Combining Internal and External Parallelization to Solve MIPs on Supercomputers

The Ubiquity Generator (UG) is a general framework for the external parallelization of mixed integer programming (MIP) solvers. It has been used to develop ParaSCIP, a distributed memory, massively parallel version of the open source solver SCIP, running on up to 80,000 cores. In this paper, we present a first implementation of ParaXpress, a distributed memory parallelization of the powerful commercial MIP solver FICO Xpress. Besides sheer performance, an important difference between SCIP and Xpress is that Xpress provides an internal parallelization for shared memory systems. When aiming for a best possible performance of ParaXpress on a supercomputer, the question arises how to balance the internal Xpress parallelization and the external parallelization by UG against each other. We provide computational experiments to address this question and we show preliminary computational results for running a first version of ParaXpress on 6,144 cores in parallel.

Evaluation of Performance Robustness of a Gas Turbine Cogeneration Plant Based on a Mixed-Integer Linear Model

In designing energy supply systems, designers are requested to rationally determine equipment types, capacities, and numbers in consideration of equipment operational strategies corresponding to seasonal and hourly variations in energy demands. However, energy demands have some uncertainty at the design stage, and the energy demands which become certain at the operation stage may differ from those estimated at the design stage. Therefore, designers should consider that energy demands have some uncertainty, evaluate the performance robustness against the uncertainty, and design the systems to heighten the robustness. Especially, this issue is important for cogeneration plants, because their performances depend significantly on both heat and power demands. Although robust optimal design methods of energy supply systems under uncertain energy demands were developed, all of them are based on linear models for energy supply systems. However, it is still a hard challenge to develop a robust optimal design method even based on a mixed-integer linear model. At the first step for this challenge, in this paper, a method of evaluating the performance robustness of energy supply systems under uncertain energy demands is proposed based on a mixed-integer linear model. This problem is formulated as a bilevel mixed-integer linear programming one, and a sequential solution method is applied to solve it approximately by discretizing uncertain energy demands within their intervals. In addition, a hierarchical optimization method in consideration of the hierarchical relationship between design and operation variables is applied to solve large scale problems efficiently. Through a case study on a gas turbine cogeneration plant for district energy supply, the validity and effectiveness of the proposed method and features of the performance robustness of the plant are clarified.

ParaXpress: An Experimental Extension of the FICO Xpress-Optimizer to Solve Hard MIPs on Supercomputers

The Ubiquity Generator (UG) is a general framework for the external parallelization of mixed integer programming (MIP) solvers. In this paper, we present ParaXpress, a distributed memory parallelization of the powerful commercial MIP solver FICO Xpress. Besides sheer performance, an important feature of Xpress is that it provides an internal parallelization for shared memory systems. When aiming for a best possible performance of ParaXpress on a supercomputer, the question arises how to balance the internal Xpress parallelization and the external parallelization by UG against each other. We provide computational experiments to address this question and we show computational results for running ParaXpress on a Top500 supercomputer, using up to 43,344 cores in parallel.