As WWW service grow up rapidly and becomes the most popular information system on the internet, web site owner invest heavily to improve the performance of web server. And, because of the 'Server Farm' architecture comes real, web server performance breaks the limit of single server, which directly cause server performance improvement. All these factors raise the cost of website performance measurement program to catch up server performance. Besides, because of dynamic webpage and database linkage are applied widely in production environment. HTTP requests (for example, sessions) with state and identifications increase rapidly. Traditional Web Benchmarking methods are out of date and not supporting these new transaction models. Also, traditional benchmarking methodologies only provide the max/average values for overview, which can not properly describe the performance of the websites with massive dynamic webpages. All these problems shows the traditional benchmarking methodologies are not sufficient for today's technologies.
In this paper, we have designed and implemented a configurable and cost-effective website performance measurement program against these problems. We introduced the concept of 'workload', and pre-design detailed HTTP request table, where time, contains, and HTTP commands to use can be assigned. Also we use configurable and replaceable design of open modules and divide the major modules into workload generator, load generator and report generator. These modules can be even separated into three independent programs, which makes this benchmark program become more flexible and adaptive to fit new technologies without adjusting the kernel. We also introduced the concept of 'Virtual User' to describe a real user behavior. We could keep the HTTP state and identification by automatic reply via cookie and assigning user identification in the same process. To increase the efficiency, each load generator can do self-diagnostics and quantify the measurements, and properly reassign the workloads by the value returned by the system. These makes every load generator do everything it could, and not been halted by the low-speed machine. And we can also prevent the mis-measurement by overloading. From the result of the experience, our design can describe the web server performance and load changes by time. We can also compare it to request category and URL, to show the root of causes and time basis to administrator. Overall, our web benchmarking methodology shows the strength to traditional web benchmarking methods.
| Identifer | oai:union.ndltd.org:NSYSU/oai:NSYSU:etd-0829101-205016 |
| Date | 29 August 2001 |
| Creators | Liang, Ming-Chang |
| Contributors | Kun-Jui Lu, Shyan-Ming Yuan, CHUNG-PING CHUNG, Sy-Yen Kuo, Chu-Sing Yang |
| Publisher | NSYSU |
| Source Sets | NSYSU Electronic Thesis and Dissertation Archive |
| Language | Cholon |
| Detected Language | English |
| Type | text |
| Format | application/pdf |
| Source | http://etd.lib.nsysu.edu.tw/ETD-db/ETD-search/view_etd?URN=etd-0829101-205016 |
| Rights | unrestricted, Copyright information available at source archive |
Page generated in 0.0025 seconds