Future Considerations

• Join Time
  It would be interesting to determine the hot page quantum for which enough requests could be made to a popular server to see significant improvements in server performance.
• Mappings
  Some reasonable mapping of pages and servers to addresses and ports should be explored. It seems improbable that this could be accomplished without some type of dynamic arbitration. The other issue is how will a client know which address to listen to prior to sending the HTTP request. One solution, similar to SDR, would be to use a single multicast address for servers to broadcast the host and page mappings.
• Not Hot Costs
  While popular servers could potentially see lower costs of serving, the wait time for pages which are not hot may be prohibitive for clients. It would be interesting to see what percentage hot pages are of all traffic, and on all servers.
• Optimizing HTTP ACK Pause Time
  Further investigation of the pause time introduced on the server at the end of multicasting would be interesting to evaluate versus the static pause time used currently. This needs to be a function of maximum round trip time and the number of nodes. Some kind of a function that maps these two parameters into time needs to be explored.