|M.Sc Student||Sedan Boaz|
|Subject||Statistical Process Control of Client-Server Software with|
Application to Internet Web Servers
|Department||Department of Quality Assurance and Reliability||Supervisors||Dr. Yefim Haim Michlin|
|Professor Dov Ingman|
Computing servers (Internet servers) are now widely used around the world. These servers are related to all aspects of our lives, in fields ranging from information and communication to shopping and entertainment. The reliable and continuous operation of these servers is a vital necessity to maintain our current way of life. However, many of these servers cease to function all too frequently- as a result of intended attacks, software errors and faulty hardware. The purpose of this research work is to propose a method for increasing the reliability of the servers by means of real-time, on-line monitoring of the servers, and detecting these undesired situations before they become fatal.
The novelty of the work lies in a new approach to web server software monitoring called the Generic Cross Correlation Process Control (GCCPC). The concept behind this technique is that a web server is portrayed as a process that can be controlled. Whereas the words process and control do not relate to computer sciences but rather to the production line, where a process is a sequence of operations that takes input material and generates output material, and control is defined as containing the output error within some tolerable limits.
Using tools from statistical process control we demonstrate that a system can be monitored and valuable information may be extracted and used to identify error conditions well in advance of a critical fault. We use methods of simulation to demonstrate the abilities of the proposed system under real-world conditions and show the simplicity by which the system may be integrated into an existing environment. A large laboratory experiment has been set-up to provide the required data to support this work.
The results obtained have shown that the method is likely to exceed in performance most known methods for early server fault detection, while being technically simple to implement and cheap to install.