Cardelini et al [9] classify web server architectures based on the entity which distributes the incoming requests among the servers in four classes of methods. Some of the methods in each category use feedback based algorithms and some use non-feedback algorithms as discussed in [1]. So we can categorize the request distribution mechanisms based on entity that routes the request as follows:
Last mechanism, Anycasting does not involve any explicit routing by web server system, but is automatically done as part of IPv6 protocol by internetworks [8].
These mechanisms offer transparency at various levels: manual selection offers no transparency because URLs are different; Client and DNS-based mechanisms may offer URL level transparency, i.e. URL is same but resolved IP addresses may be different; Dispatcher based approaches usually offer IP address level transparency, i.e. even resolved IP address is also same.
Some mechanisms are geographically scalable, i.e. cluster of servers can be either in LAN or WAN. Some approaches are fault tolerant and highly available but others are not. Some approaches require replication of whole web-site, while others allow partial replication.