We designed and implemented a test bed for evaluation of load balancing strategies for distributed web server systems. This test bed is quite flexible and new policies can be compared with already existing policies very easily. This test bed will help in understanding trade offs and impact of different parameters on a distributed web server system.
In our thesis, we proposed an adaptive and dynamic policy for server selection and request distribution for a very large website. This DWSS can be deployed with current infrastructure and protocols in use. This architecture is scalable and fault tolerant too. In short, it meets all goals mentioned in design section.
We modified IP packet forwarding method to rewrite only incoming IP packets using shared common IP packets. This can be implemented totally at application layer with divert socket and IP firewalling support, since packets from clients are much shorter, even at application layer there is less overhead as compared to rewriting reply packets which was used in earlier proposed request distribution mechanisms.
From results obtained, we can conclude that our architecture will give better results when clients accessing a particular site are spread in different geographical regions and they are far away from each other. Our architecture is geographically scalable as well as fault tolerant for new incoming requests. Our architecture achieved its main goal of minimizing response time perceived to client.