| Author: | Xiangping Chen |
| Advisor: | Prasant Mohapatra |
| Email: | chenxia1@cse.msu.edu; http://www.msu.edu/~chenxia1 |
With the success of the Internet and the World Wide Web, the variances in Web application types and their service quality requirements have been also increasing. On the other hand, the bursty nature of Internet server workload degrades the system utilization and results in unpredictable response time to client requests. Differentiated service has been proposed as a potential solution to provide statistically predictable quality of services (QoS) and is expected to be supported in the next generation Internet. However, a FCFS scheduling server may defy the efforts from a service differentiating Internet. Furthermore, the fast growth in e-commerce demands service classification from its servers to meet the goals of sales and services. Service Differentiating Internet Servers (SDIS) is designed to provide QoS assurance in terms of response delay and throughput to client requests. A SDIS realizes predictable QoS by categorizing requests into different priority groups based on task types and client identifications. Admission control, scheduling, task assignment policies, and overload control issues are considered with respect to priority groups and task types. Through analytical model and simulation study, we have shown the feasibility and performance benefits of SDIS in various aspects. Experimental results using real web server traces have proved that a service differentiating Internet server provide significantly better services to high priority tasks compared to a traditional web server under high system utilization. Further improvement in performance can be improved by efficient admission control and overload management.