Performance Optimization of Complex Systems
Abstract
The research results in this report are based on an effort to develop systematic techniques for performance optimization of complex systems with special emphasis on real-time methods. Recognizing the increasing importance of stochastic networks in both the civilian and military domains, explicit algorithms are sought that are scalable, distributed, asynchronous, and computationally compatible with the limited processing capabilities at individual nodes of many such networks. The main outcomes of the project are: (a) An asynchronous event driven distributed optimization framework allowing autonomous agents to cooperate toward a common goal with minimal communication among them, thus saving energy without any loss in performance. In particular, communication is limited to instants when a state estimation error function at some agent exceeds a threshold, (b) An optimization framework for systems with time-critical tasks in which hard real-time constraints are guaranteed to be satisfied. At the single node level, an efficient solution procedure termed the Critical Task Decomposition Algorithm (CTDA) was developed. At the multi-node level, a Virtual Deadline Algorithm (VDA) was developed. Both algorithms are scalable in the number of tasks executed, (c) Extensions to perturbation analysis methods for gradient estimation and optimization of Stochastic Fluid Models (SFM) as abstractions of complex stochastic systems.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 2009
- Accession Number
- ADA495123
Entities
People
- Christos G. Cassandras
Organizations
- Boston University