Establishing correct/smooth and efficient workflow in modern production, transportation and business systems is important to organizational profitability and success. One way to systematically study the correctness and efficiency of the workflow taking place in the aforementioned environments, is by modeling them as resource allocation systems (RAS). In the RAS modeling abstraction, a set of transient entities, known as jobs or process instances, are executing a well-defined sequence of operations, otherwise known as process stages, by employing at each stage various subsets of the system resources.  The resources are reusable entities that are available in a finite number of copies / units, known as the resource capacity, and when allocated to a particular job instance, they are occupied by it in an exclusive manner.  Hence, two important classes of control problems arising in the operation of this type of system are as follows:

• RAS Structural Control, which ensures logically correct system behavior, so that each job executes through its full sequence of operations without human / external intervention to resolve potential conflicts / deadlocks in the ongoing resource allocation; and
• RAS Performance Control, which further coordinates / schedules the resource allocation enabled by the applied structural controller, in order to promote optimality in some performance measure (e.g., throughput or cycle time).

Traditional operations research approaches to the RAS control problem have focused on performance control, without considering the effect of low-level system characteristics addressed in the structural control problem.  However, issues related to structural control become more and more critical as many of the aforementioned systems are migrating to more highly automated operational modes. Furthermore, our past findings indicate significant differences in the characteristics of optimal performance control approaches when one considers those system characteristics relating to structural control. Motivated by these remarks, this research program seeks to develop formal approaches that will allow (i) the systematic characterization of the optimal performance control policies that are appropriate for the structurally controlled RAS models, as well as (ii) the development of effective and computationally efficient approximations, since the optimal policies themselves are expected to be computationally intractable.

Contact: spyros@isye.gatech.edu