NEOS Server

NEOS Server
Name	NEOS Server
Developer	Argonne National Laboratory, University of Wisconsin–Madison
Released	1996
Programming language	Perl (programming language), Python (programming language), C (programming language)
Operating system	Unix-like, Linux, Microsoft Windows
Genre	Optimization server

NEOS Server is a client-server system for solving numerical optimization and computational problems via remote submission to a pool of solvers hosted on centralized servers. The project provides network-accessible interfaces that connect users and applications to optimization AMPL-compatible, GAMS-compatible, and other solver back ends, enabling distributed access to linear, nonlinear, integer, stochastic, and large-scale solvers. NEOS integrates with batch scheduling systems and high-performance computing resources to offer remote submission, monitoring, and retrieval of results.

Overview

The NEOS Server offers a web-based and programmatic gateway that mediates between clients and solver daemons, supporting protocols for job submission, queuing, and result delivery. It interfaces with modeling systems such as AMPL, GAMS, MATLAB, R (programming language), and Python (programming language) front ends, and connects to solver packages including CPLEX, Gurobi, IPOPT, SNOPT, and BARON. The architecture enables integration with compute resources managed by systems like SLURM, PBS (software), and HTCondor, and can route jobs to supercomputers at facilities like Argonne National Laboratory, Lawrence Berkeley National Laboratory, and university clusters.

History and Development

Development began in the mid-1990s with collaboration among researchers at Argonne National Laboratory and University of Wisconsin–Madison to provide remote optimization services to the research community. Early milestones included adoption by modeling language communities around AMPL and GAMS and incorporation of solver interfaces for commercial packages such as CPLEX and academic codes like MINOS. Over time NEOS expanded support for stochastic programming researchers using frameworks from Stochastic Programming Society efforts and interfaced with emerging solvers from groups at Stanford University, Massachusetts Institute of Technology, and University of California, Berkeley. Major updates paralleled advances in middleware and scheduling from projects at Lawrence Livermore National Laboratory and National Center for Supercomputing Applications.

Architecture and Components

The NEOS Server uses a modular design composed of a submission front end, a scheduler/queue manager, and solver stations or execution daemons. Front ends include web servers compatible with Apache HTTP Server deployments and SOAP/REST interfaces influenced by standards promoted by World Wide Web Consortium. The scheduler can interface with resource managers such as SLURM and PBS (software), and uses wrapper scripts often written in Perl (programming language) or Python (programming language) to manage solver invocation. Solver stations host binaries from vendors like IBM (for CPLEX), Gurobi Optimization, LLC (for Gurobi), and academic projects from Princeton University, Cornell University, and Georgia Institute of Technology.

Supported Problem Types and Solvers

NEOS supports a broad taxonomy of optimization problem classes: linear programming instances solvable by CPLEX and Gurobi; mixed-integer programming addressed by CBC (Coin-or branch and cut), SCIP; nonlinear programming tackled by IPOPT, SNOPT; global optimization handled by BARON and Couenne; and stochastic programming via solvers developed in collaborations involving University of Iowa and INFORMS communities. The server also accepts conic programs, quadratic programs, complementarity problems, and systems of equations studied at institutions such as Princeton University and Carnegie Mellon University.

Usage and Interfaces

Users submit jobs through a web portal, e-mail interfaces, and programmatic clients that follow protocols similar to standards in projects at National Institute of Standards and Technology and Open Grid Forum. Integration libraries exist for MATLAB, R (programming language), Python (programming language), and modeling environments like AMPL and GAMS, and workflows commonly connect NEOS to continuous integration tools influenced by practices at GitHub and GitLab. Authentication and access control have been aligned with institutional identity providers exemplified by InCommon and campus federations associated with Internet2.

Performance and Benchmarking

Performance characterization of NEOS deployments often references benchmark suites and competitions organized by communities such as MIPLIB and CUTEst, and leverages benchmarking methodologies from Benchmarking (computing) initiatives at National Renewable Energy Laboratory and university research groups. Comparative studies examine solver performance on representative instances from repositories like COIN-OR and evaluate throughput and latency under load using cluster technologies researched at Lawrence Berkeley National Laboratory and Sandia National Laboratories.

Impact and Applications

NEOS has enabled research, education, and industrial applications by lowering barriers to access high-quality solvers for users affiliated with universities like University of Wisconsin–Madison, national laboratories such as Argonne National Laboratory, and companies using optimization for logistics, energy, and finance problems studied at MIT and Stanford University. The service has supported reproducible computational experiments cited in publications at conferences sponsored by SIAM, INFORMS, and IEEE, and it has facilitated collaborations across academic groups including Columbia University, Yale University, and University of California, Los Angeles.

Category:Optimization software