M. D’Apuzzo*, M.L. De Cesare**, M.R. Maddalena**, M. Marino**, G. Toraldo** Collaborators: S. Cafieri*, V. De Simone*, D. di Serafino*, E. Sacchettino* * Second University of Naples **University of Naples Federico II Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 ESOPO: an Environment for Solving Optimization Problems Online 1 Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 http://www.firb_lsno.unina.it 2 Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 3 • ESOPO aims and structure overview • Relevant features of ESOPO • Perspectives and future enhancements Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Overview 4 to provide a unifying framework containing the optimization software produced by people working in the MIUR FIRB project, in order to interact in the software development, testing and evaluation processes Several issues •Shared software classification criteria •Common linear algebra kernels •Common optimization subproblems •Standard software documentation •Shared test problems •Similar input formats Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Early motivation for ESOPO 5 to be a web-based environment for solving optimization problems and for evaluating and comparing the performance of optimization software Several issues Software integration procedure Robustness and reliability Preprocessing and presolving stages Drivers to the solvers for using common problem modeling languages Minimal input effort Testing process Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Current ESOPO’s ambition 6 to be a web-based environment for solving optimization problems and for evaluating and comparing the performance of optimization software Several issues Interactive procedure for solving a problem Interactive choice of a solver Dynamic interfaces for using the solver Automatic selection of test problems based on the type of considered instance Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Current ESOPO’s ambition 7 MAIN ACTIONS • collect, integrate and make available the optimization software produced in the MIUR-FIRB Project, toghether with some well established software (Lancelot, KNITRO, Mosek, ...) • supply the solvers with drivers for the most common problem modeling languages and with graphical interfaces for a friendly usage • provide suitable collections of test problems and up-todate tools for evaluating and comparing optimization software Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 ESOPO project 8 ESOPO: SOLVE problem user provided or selected from collections solution ESOPO: PERFORMANCE EVALUATION set of problems performace evaluation user provided or profiles selected from collections Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Main ESOPO abilities 9 • Solvers ..... • Drivers ESOPO Solvers job execution results answer • Users database • Software and Problems database • Interfaces for choosing solvers and for submitting problems • Tools for job queuing ESOPO Server request ..... Clients (browsers) Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 ESOPO architecture client-server design 10 • software integration process • interactive procedure for choosing a solver and for solving a problem • close integration of solvers and test problems • integration of the solving tools with the benchmarking tools Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Relevant features 12 Integration and management of the Software (authors are only request to submit the code!) Step 1: Classification into ESOPO Example: SDBOX (solves general bound constrained nonlinear optimization problems using a derivativefree method) OP: local; OF: general; CO: bounds; DR: none; CVX: no; STR: dense Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Relevant features 13 Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 14 Integration and management of the Software Step 2: Development of drivers to the solver • Make its use through dynamic web pages easier • Provide interfaces to AMPL and SIF modeling languages • Reduce as much as possible the number of input parameters • Perform the testing process • Supply some extra features to the solver Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Relevant features 15 Interactive procedure for solving a problem (problem oriented and independent of the computing engine) Step 1: Specification of the problem web interface that allows the user to supply information about the problem to be solved Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Relevant features 16 Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 17 Interactive procedure for solving a problem Step 2: Selection of a solver web interface that lists all solvers available for the problem Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Relevant features 18 Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 19 Interactive procedure for solving a problem Step 3: Choice of the input format tailored interface for the selected solver (automatically generated) allowing the users to choose the input format among those accepted by the solver Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Relevant features 20 Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 21 Interactive procedure for solving a problem Step 4: Submission of the problem specific interface consistent with the user’s choice for the input format (automatically generated) that allows the user to provide the problem data and the values for the input parameters Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Relevant features 22 Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 23 Close integration of solvers and test problems A set of test problems that the software is able to solve is automatically selected Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Relevant features 24 Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 25 *************************************************************************** * * * Output report from ESOPO * * * *************************************************************************** SOLVER: SDBOX PROBLEM: BIGGSB1 from CUTEr collection VERSION: AMPL # # # # # # # Source: M. Batholomew-Biggs and F.G. Hernandez, "Some improvements to the subroutine OPALQP for dealing with large problems", Numerical Optimization Centre, Hatfield, 1992. SIF input: Ph Toint, April 1992. classification QBR2-AN-V-V NVAR = 5000 INPUT PARAMETERS: TOL = 10e-6 - MAXITER = 1000000 RESULTS: NIT = 181158 NFEVAL = 544749 FVAL = 0.015003 Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Execution report 26 Interactive procedure for evaluating and comparing the performance of optimization software The solving and benchmarking stages are integrated in ESOPO Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Relevant features 27 28 1 2 3 4 5 6 7 8 9 10 1 1 1 2 Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Performance evaluation report Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 ESOPO contents 29 • to add more solvers also in areas not currently covered • to improve the interaction between users and ESOPO • to provide other metrics for the performance evaluation Convegno Progetto FIRB LSNO – Capri 19/20 aprile 2007 Future developments 30