An algorithm for resource management in large systems

This work studies the application of decentralization to the problem of Resource Management (RM) in Large operational Systems, where a very large number of Activities share limited resources. Our emphasis is on the feasibility aspect of the problem, that is, of taking a large system and keeping it operational in the face of changing characteristics of the Activities. Motivation for our study is provided by the RM task in a large FIAT warehouse. The size of the problem, and the ill-behaved resource-usage functions, make standard techniques unsuitable. However, by replacing the feasibility problem by a suitable "Artificial" optimization problem, we can use Lagrange Multipliers to provide a simple solution through decentralization of decisions. A Theorem is presented giving simple conditions for The existence of optimal multipliers for the Artificial Problem. Algorithms to solve the RM problem are also given, having proveable convergence properties, and quadratic convergence rates. (Our theorems are proved without the usual strict convexity conditions.) The design of a practical RM system is described with examples from the system designed for the FIAT warehouse.