An agent model for computer performance enhancement | |
| Author | Poonphon Suesaowaluk |
| Call Number | AIT Diss. no.CS-05-03 |
| Subject(s) | Computer systems Intelligent agents (Computer software) |
| Note | A dissertation submitted in partial fulfillment of the requirements for the degree of Doctoral of Technical Science in Computer Science and Information Management Program, Schoool of Advanced Technologies |
| Publisher | Asian Institute of Technology |
| Series Statement | Dissertation ; no. CS-05-03 |
| Abstract | Agents may represent living humans, machines or entities of various kinds. There may be agents representing agents themselves. They negotiate with one another to maximize their interests based on their beliefs and aspirations. The networked world can therefore be viewed as a world of agents where every agent is in principle aiming to negotiate with every one else. Recognizing the intractability of theoretical formulations, a practical problem of load sharing among machines from these perspectives is addressed. Although negotiations manifest in many forms, the collaborative process is best suited in discussing the load sharing process. When a number of computer systems are connected over the network, the quality of service the user obtains depends not only on the performance of the machine he is interacting, but also on those not directly involved with this user, as the resources are exchanged. Typically every conventional computing machine comes with a usual set of resources, such as processing power, memory, peripheral, data bases and software. It has been observed that most machines, in a university set up, for instance, are under utilized in terms of their resource consumption. Many research efforts have been reported to revitalize these idle resources to meet the demand occurring anywhere in the network. Moving and balancing the load across the interconnected machines may improve the response time, turn around time and the over all quality of service of the system. In this work, a scenario is envisaged in which agents have tasks to perform and are in a position to exchange tasks to improve their performance, and thus perhaps affect the overall performance of the system. The machine agent at low loads would aspire to obtain load to increase its contribution to the network. Again the machine having substantial load would like to shed some of its load to ensure reasonable turnaround time and response time. We have designed a utility function to reflect this kind of aspiration of the machine agent. The model initially uses two computing machines with their usual resources, and with random initial jobs to be processed. Their performance is dependent not only on absolute total load but also on how much of each machine's resources are in demand, thus giving scope for the beneficial exchange of jobs. The research describes the development of a negotiation mechanism for job exchange between machines and measurement of system performance. An analogy is drawn to the case of an exchange of letters between postal deliverymen to increase the speed and efficiency of the delivery system and similarity and differences between the two domains are noted. This research is on a two machine networked model, to investigate the paradigm of negotiation on resources sharing and a greedy method is used as a heuristic for the exchange of jobs. It is then generalized to a network of three machines. We have also investigated the operation of these concepts on the commercial domain of business, where exchange of stocks may be effected among retail outlets. The concept may be extended to multiple machines that act as agents exchanging resources and tasks, such as the World Wide Web which is a very large number of machines and resources. The agents connected may be expected to effect a prudent exchange of jobs in order to enhance individual performance and overall system performance. However, as the number of machines in the network increases, the complexity of protocol would tend to be exponential. Taking a real life situation, a "small world" concept could be used. The concept is to form a cluster of small groups of nodes within the network for interaction. However, we have not pursued this at the time of research |
| Year | 2005 |
| Corresponding Series Added Entry | Asian Institute of Technology. Dissertation ; no. CS-05-03 |
| Type | Dissertation |
| School | School of Advanced Technologies (SAT) |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Computer Science (CS) |
| Chairperson(s) | Sadananda, Ramakoti; |
| Examination Committee(s) | Huynh Ngoc Phien;Tripathi, Nitin K.;Murugesan, San; |
| Degree | Thesis (Ph.D.) - Asian Institute of Technology, 2005 |