Improving the performance of Web servers has become a critical issue in coping
with the increasing use of network based services. The critical nature of many online
transactions and distributed services mandate design of high performance Web servers
since such servers are anticipated to be the bottleneck in hosting network based services.
The research objective of this thesis was to analyze and enhance the performance of
cluster based Web server/data center based on the user level communication paradigm.
To achieve this objective, we investigated several design issues in cluster based Web
The major contributions of this thesis are summarized as follows. First, we pro
posed a coscheduled Web server model, which adopts coscheduling mechanisms between
communicating processes of Web servers to reduce the response time of requests. The
round trip time is reduced, the communicating processes of the requests are scheduled by
the operating system as soon as possible. Also, the round trip time is reduced when the
coscheduling algorithms reduce the context switches of processes. We analyzed four dif
ferent Web server models; PRESS over VIA, coscheduled PRESS with DCS, coscheduled
PRESS with DCS and blocking, and Adaptive, in 16 node and 32 node configurations