Globule > publications

Decrepol: a Decentralized Replication Policy for Web Documents

Author: Paul Zijlmans
Source: Masters thesis, Vrije Universiteit, August 2006.


The current size of the Internet makes it impossible to host major Web sites on a single server. Replication can improve both the availability and the performance of a Web hosting service. Current replication policies heavily rely on the origin server. Therefore, this thesis presents Decrepol, a decentralized replication policy for Web documents that assumes the origin server to be unavailable most of the time. First, we present a replication policy that achieves full replication and is structured following epidemic protocols. With this policy we are able to spread notifications across 128 replica servers in about 5 gossip rounds. Next, we present a decentralized replication policy that allows for controlled partial replication. We create two versions of the partial replication policy. The first one is based on epidemic protocols and is an extented version of the full replication policy. However, in this policy locating replicas becomes a problem. Therefore, we also present an improved unstructured version that facilitates locating content. With this version, inserting and retrieving documents requires traversing about three hops. The second version is based on structured peer-to-peer systems. Inserts take about four hops, retrieves only two hops if we cache document searches. We show that it is possible to create an unstructured partial replication policy that performs almost identically to a structured one and thus can be an interesting alternative.

Keywords: Replication policy, Web hosting, Peer-to-peer, Decentralization.


* The thesis, in PDF (558,591 bytes).

Bibtex Entry

  author = 	 {Paul Zijlmans},
  title = 	 {Decrepol: a Decentralized Replication Policy
                  for Web Documents},
  school = 	 {Vrije Universiteit},
  address = 	 {Amsterdam, The Netherlands},
  year = 	 {2006},
  month = 	 aug,
  note =         {\url{}}