Work

Load balancing control of a server network cluster

Year: 2013
Type: dissertation
Abstract: Faced with increasing network services and number of users, requests to the servers\nat those sites has signi cally skyrocketed. Moreover, most of these servers need to\nrun twenty-four hours a day, 7 days a week with a high reliability and availability.\nConsequently, the tremendous growth of the Internet has led the requirement of multiserver\nstructures in order to deal with these e ectiveness issues. A much higher processing\npower may be provided by a set of computational elements than by a single one, even if it\npresents a powerful capacity. Additionally, the global system throughput may greatly increase\nby using properly these architectures. However, to make an e cient server network\nis a di cult task. This is the main goal of this dissertation.\nE ciency may be increased if server network works in cooperation, distributing the\nload. This is known as "load balancing". This technique may make that network is robust\nand e cient, that means, overload are avoided at the same time that system resources are\nwell exploited.\nAssuming that all servers present a same architecture, a deterministic dynamical model\nis designed and a distributed control law inspired by consensus theory is developed. In\nthis way, the closed-loop ensures load balancing as well as asymptomatically stability.\nMoreover, thanks to decentralized control, computational load is reduced and scalability\nis possible. On the other hand, an attraction domain is estimated to ensure positive rates.\nThe e ectiveness of the distributed control is validated by simulations in Matlab &\nSimulink and Network Simulator 2.
Language: English
Open Access status: green
OpenAlex ID: W2160262121
Cites: 29
Cited by: 2
Related to: 10