This section presents an extension of the LARD policy that works efficiently in the presence of HTTP/1.1 persistent connections, when used with the back-end request forwarding mechanisms presented in the previous section.
Both the front-end relaying mechanism and the TCP multiple handoff mechanism allow requests to be distributed at the granularity of individual requests. As such, they do not place any restriction on the request distribution policies that can be used. In particular, the LARD policy can be used in combination with these mechanisms without loss of locality.
The back-end forwarding mechanism, on the other hand, does place restrictions on the distribution policy, as it mandates that a connection can be handed off to a back-end node only once. If requests arrive on a persistent connection that cannot or should not be served by that back-end node, the policy must instruct that back-end node to forward the request to another back-end node.
We have developed an extension of the LARD policy that can efficiently distribute HTTP/1.1 requests in a cluster that uses the back-end forwarding mechanism. The following subsection briefly presents the standard LARD strategy. Then, we proceed to present our extension.