Towards Self Managing Distributed Systems Capable of Resolving Trade-Offs Among Competing Performance Criteria
Vladimir V. Marbukh
Since typically network performance is characterized by multiple competing criteria, the network management requires resolving the corresponding trade-offs. The problems include finding the Pareto optimal frontier in space of these criteria and selecting the desired operating point on this frontier. Mathematically, finding the Pareto optimal frontier can be framed as a constrained optimization problem. Solution to this problem can be expressed in terms of the Lagrange multiplies, which characterize the sensitivity or the price of one competing criteria or constraint with respect to another. Once these prices are flooded throughout the network, the agents have the necessary and sufficient information for making the optimal decisions on amount and mixture of the resources to request. These decisions may relate to flow control, routing, etc. Finding optimal operating point on the corresponding Pareto optimal frontier in a distributed environment is equivalent to aggregate utility maximization, where the aggregate utility is the sum of individual utilities of all the elastic users. Importance of the concept of an elastic user capable to adjust its rate in response to the external stimuli by maximizing its individual net utility has been articulated in [I], and the corresponding aggregate utility maximization framework has been proposed by in. . This paper advocates using the aggregate utility maximization framework for balancing a wide range of conflicting requirements of the elastic users/contracts, capable of adjusting not only its transmission rate, but also a wide range off burstiness and quality of service parameters as well as willingness to expend resources.
Proceedings of Sixth Workshop, 2004
June 12-13, 2004
Mathematical Performance Modeling and Analysis (MAMA)