Please use this identifier to cite or link to this item: http://hdl.handle.net/10397/85023
Title: Design and implementation of cluster computing system
Authors: Cheuk, Wai-kong
Degree: M.Phil.
Issue Date: 2001
Abstract: In recent years, the emergence of Internet, local area network, cluster computing and parallel processing have dramatically shortened the communication time between processing units as well as the other computational resources. Cooperating execution between processes located in different machines connected only by a communication link becomes possible. The building of computer applications on such platform for practical and functional operation becomes possible. However, such advancement has brought new problems. Firstly, experiment shows that when there are additional processes put to execute in a time-sharing machine, the performance of processes within the machine degrades. Stemmed from the lacking of guarantee in computation power allocation, this phenomenon seriously affects the performance of execution on any timely related application. Applications such as video player and sound player suffered as they are having their execution progresses strongly related with time. Introduction of additional processes to the system will lower the share of computation power for the existing processes; missing frame will be resulted. A guaranteed delivery on allocated computation power is therefore required to solve the problem. To achieve this, we must have an invariant and explicit representation of the allocated computation power. In this thesis, a novel approach for representing computation power is proposed. The computation power is represented in physical time form, with attributes describing the delivery rate - in terms of the processing time in a specific period together with the start time and end time. It is possible to generate two different types of computation power pattern with this representation - the power form and energy form. The power form describes a periodic consumption of power, which can be viewed as the amount of power to be delivered in a period. While the second form - energy - is used to describe power consumption in burst form. As there is a guarantee of service in form of time, the system cannot guarantee service to another process at the same time. Therefore, an admission control system is developed to settle the dispute. Finally, in addition to the representation itself, an actuation mechanism is developed in this work to allow the guaranteed services to operate without fault. Another problem encountered in the current operating system is that, when multiple processes with different execution rate are required to execute cooperatively in a system, the proper execution rate of the higher one may not perform correctly. This is the problem of priority inversion. This phenomenon occurs when multiple processes in cooperation executed with different computation power requirement. During the execution of a process with higher computation power requirement, it may need to synchronize its progress or communicate with the lower one and the process blocks until the synchronization / communication is finished. As the process with lower power requirement will execute slower than the higher one does, the higher power process has to forfeit its chance of execution because its synchronization/communication operation with the lower power one has blocked itself. Degradation in performance is resulted. The occurrence of this problem owes to the lack of capability in making controllable transfer in computation power between processes; it is not possible to transfer the computation power relinquished by a process to another process directly. In order to facilitate the controllable transfer, it is necessary to setup the relationship between owner and transferee. In this thesis, this problem is addressed by the introduction of the concept - process group. A process group is a group of processes led by a leader process; normally, it will be scheduled for computation power according to its specified delivery rate. However, when it relinquishes its computation power, the power will only be given to the process within its group, thus, a controllable computation power transfer is achieved.
Subjects: Parallel programming (Computer science)
Hong Kong Polytechnic University -- Dissertations
Pages: xii, 158 leaves : ill. ; 30 cm
Appears in Collections:Thesis

Show full item record

Page views

50
Last Week
0
Last month
Citations as of Mar 24, 2024

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.