The rapid advancements in technology have caused the complexity of computer systems, networks and applications to grow at an explosive rate.Performance analysis, modeling, and optimization have played a critical role in understanding fundamental problems in the design, development, management, and planning of such complex systems and services. IBM Research has a long history of making fundamental contributions in these areas. Much of our research has also contributed to successful IBM products and services.

A key focus area of our research is on improving different aspects of server performance. Significant pieces of systems work have led to technology for high-volume Web sites, including new techniques for efficiently generating and serving dynamic content. These techniques have been successfully deployed to improve performance at a number of high-volume Web sites such as those for the 1996, 1998 and 2000 Olympic Games. We have further developed new techniques for load-balancing requests among multiprocessor Web sites, and several of these techniques have been incorporated into IBM's Network Dispatcher product. Significant work has also been done on Web and proxy server acceleration, as well as in operating system extension technology and kernel components for high-performance network servers.

Another focus of research concerns the analysis and modeling of workload characteristics and performance in different environments. This is key to gaining insights into the complex workload characteristics found in practice and the impact of such complexities on different measures of performance (e. g., quality of service, scalability, various workload and response-time characteristics, availability, and reliability). Our research in this and related areas has made
significant theoretical contributions in various fields including queuing theory, stochastic models and analysis, forecasting and scaling, and simulation. The derivation of these mathematical methods and results has been in turn exploited to develop practical methods and results for different aspects of system performance analysis, including bench-marking, capacity planning, and forecasting of workload and performance processes.

The control and optimization of performance and other measures in complex computer-based environments is also a key research focus. This research addresses fundamental problems across a wide range of areas, including control of queuing networks, dynamic scheduling, load balancing, admission control, capacity planning and inventory management, off-line and on-line algorithms, quality of service, and service level management. The derivation of methods and solutions to these problems has also been exploited in practice by incorporating the orresponding results in system and application components, such as resource management and scheduling,
and performance management.

Tools and methods that support measurement, monitoring, and performance analysis are another important focus area of research. As a specific example, one tool we developed analyzes end-to-end Web page performance. For each page component, this tool decomposes its delivery
time into that taken to: resolve domain name(s), establish socket connections, request the page component; and retrieve it. Using this information, Web page designers and site operators can make informed decisions regarding the performance of their page designs.

Another performance tool under development automates the assessment of web page performance data, making recommendations for areas of investigation to improve Web page retrievals and showing how proposed changes might affect response time. We are also developing technologies that provide continuous monitoring of Web page retrieval requests as well as complex e-business transactions for service level management.

Please contact Paridhi Verma to obtain copies of the Computer Science Brochure