Processes typically consist of a prestructured set of activities performed by different resources (people, machines) subject to time and cost constraints. This paper demonstrates an algorithm to determine the cost-optimal allocation of resources to activities, satisfying the given time and activity flow constraints.

In contrast to other approaches focusing on central coordination, this work approaches the problem from a distributed virtual enterprise point of view. Three types of intelligent software agents then represent the overall process, individual resources, and subprocesses corresponding to logically connected parts of the whole process (for example, representing the activity of a single manufacturing machine or a continual string of activities performed in the same department).

Using time Petri nets to mathematically represent the domain (including agents), the authors present four small algorithms to be carried out by the agents in order to solve the resulting workflow composition problem. An order fulfillment process serves as an excellent nontrivial example, where all mathematical definitions and steps of the algorithm are presented in extensive detail with lots of enlightening figures. As a side effect, the algorithm also allows the generation of state-dependent task lists for the resource agents. True to the virtual enterprise paradigm, the algorithm also uses the contract net protocol (CNP) for interagent communication.

Apparently, the authors have implemented their algorithm on the Java Agent Development Environment (JADE) platform. However, besides depicting some (trivial) screen shots and presenting the highly superficial implementation architecture, this is not further explored in this paper.

I especially liked the distributed, agent-based approach to solving this optimization problem and the lucid style of the paper. I definitely recommend it to anyone in industry research or academia working on the forefront of automatic workflow composition or optimization.