An effective self-adaptive bi-directional multi-layer genetic algorithm for resource and user allocations in networks
Expert Systems with Applications
Transportation, telecommunication, electricity supply, and production–distribution systems all involve the construction and management of highly sophisticated networks. Without properly designing such networks, none of these systems work effectively. Network design problems, as the core of network-based decision making systems, can be used to model the technological design of networks. They are widespread and have many industrial applications. In telecommunication, usually an infrastructure is sought, as a combination of possible links and switches, which can best maintain a given level of data traffic between a number of servers and devices. This paper presents a solution to a problem in which in every period of time, the specification of some of the nodes in network may need to be altered. In the current procedure, all arcs and nodes are fixed and only the node parameters need to change. However, the presented solution strategy, because of its multi-layered characteristic, can be easily extended to the cases where arcs and nodes are also subject to change. The solution procedure has been tested on the cloud-based Internet Café problem. In this problem, different types of video games can use different amount of RAM, CPU, GPU, and video-RAM resources, with demand for each type of game varying dynamically in a specified period, e.g., over a day. Different servers provide these resources to dumb-terminals which act essentially as input–output devices, including display monitors, for games running on the servers. Each server needs to provide the necessary resources for the games currently played on all terminals served. In this way, depending on the demand of each terminal, at each period, the terminal's capability can change dynamically, and it will act as a different virtual machine in different periods. Hence, the requirements of the Internet Café infrastructure change dynamically over time. The solution procedure has been tested on a series of generated large instances, and the results are very promising. For a total of 9 cases tested, the optimal solution is obtained, in a matter of minutes, for 3 cases. For an additional 2 cases, the solution obtained is within less than 2% of the optimal solution. The results are also reported for 3 additional test cases for a problem with over 13,000 users and 4,000 seats, demonstrating the procedure capability in handling extra-large problems.
Open Access Status
This publication may be available as open access