Our simulations showed that push-caching complements client-caching in addressing the scalability problems plaguing the World Wide Web, but we must also acknowledge that push-caching is most effective when the full Internet topology is known. Geographical distance can help to predict topology, but the more accurate hop-based metric consistently out-performed the miles-based metric by at least 25%.
We have shown that push-caching is a promising approach to solving Web scalability problems on the Internet, but we hope that we have also shown that push-caching is generic enough so as to be applicable to other large-scale distributed systems. As more and more computers become networked the need for efficient, scalable, and dynamic load-balancing and replication algorithms will increase. Push-caching is promising because it meets all of these criteria without requiring network architecture extensions or human intervention. We hope that the ease with which a server-initiated caching architecture may be deployed will lead to its rapid acceptance and integration into today's large-scale distributed systems.