engleski

Service optimization by self-adaptation in the origination and termination part

Abstract - This paper presents an unconventional approach to designing routing properties in and outside network nodes and other distribution points. The work has been motivated to detect additional capacities hidden inside the information about different abilities and properties of network nodes, links and routing points, particularly because they usually act in distributed environments. Potential redundancy has been detected in telecommunication network consisting of multiple nodes, specialized for routing jobs between the clients and service units. Two communication sites (points) have been subject of our consideration in the given client-server environment. These points belonged to the originating and terminating part of service. Two independent sets of simulation were done for originating part (client) and terminating part (server units) of service. Selection criteria were based on continuous adaptation during the actions. Time shift was used to simulate self adaptation capability for the selection routing points distributed in different time zones. Continuous adaptation was achieved by reinforcement Q-learning. Simulations showed that implementation of knowledge about past behavior could, in particular situations, significantly accelerate the entire service system. There was rapid adaptation and significantly improved performance for self adaptable selection compared to sequential selection in the originating and terminating part of the service.

reinforcement learning; service optimization; client-server communication

nije evidentirano