In this section we describe our Web++ client design and discuss the major design decisions. Web++ client binds logical URLs received in every HTML resource to one of the physical URLs corresponding to the replicas of the resources referenced by the logical URL. By binding a logical URL to a physical URL, the client selects the ``closest'' replica of a resource and fails over to another replica if the closest replica is not available. The current Web++ client implementation consists of approximately 18 KB of Java bytecode.