SLP-IOR: On the Design of a Workbench for Testing SLP Codes

The development of efficient theoretical algorithms and their computer implementations are intimately related activities. For testing an implementation of a numerical method the availability of the existing test problems and access to a facility for randomly generated test problems are important issues. In the field of stochastic programming a standard input format for test problems has been proposed to facilitate the communication of these problems [2]. A test problem generator for two-tage problems with complete recourse is also readily available [19]. For a collection of test problems see [20]. Despite these facts comparing performance of implementations of new algorithms with existing solvers remains a cumbersome task especially if new or modified test problems are involved. Anyway, to get an idea of comparative performance the solvers should run in the same computing environment. At the Institute for Operations Research of the University of Zurich a model management system for stochastic linear programming, SLP-IOR, is under development. One of the major design objectives for SLP-IOR was to support the testing of new solvers. The subject of this paper is to summarize the main features of SLP-IOR from this point of view. A more detailed description of SLP-IOR can be found in [16]. For model management systems in OR see [6]. For the related subjects of modeling environments and intelligent systems see [11] and [31], respectively.