We did not look at the other policies for request distribution within each cluster. Besides it, we assumed that all the servers are capable of serving all the requests. Next step would be consider partial replication on different servers within each cluster and come up with a policy at cluster level to distribute different requests to different servers to get advantage of caching at each server. Next step will be to support different quality of service to different customers or to provide better response time to clients requesting a subset of URL space, for e.g. when they visit pages related to shopping at site.
Our test bed can be further generalized to have switching at different layer of network, we have source code for symmetric and asymmetric splicing too. If support of switching at higher layers is provided, policies that make use of URL or application layer content to select server can also be emulated on our test bed and can be evaluated easily.
After integration of switching at application level, support for partial and dynamic replication can also be added to make it more comprehensive test bed. Once request distribution policies and replication strategies are in same test bed, dynamic replication policies can be explored further which are still not properly understood and explored in distributed web server systems.