CS8803D Course Reading Summaries Title: A Scalable-Content Addressable Network Problem:

advertisement

CS8803D Course Reading Summaries

Title: A Scalable-Content Addressable Network

Problem:

In Peer-to–peer file sharing systems such as Napster and Gnutella files are stored at the end user’s machines and not at a central server. Many peer to peer systems face stability issues.

Napster needs a central server to store index of all files. Gnutella floods request with a certain scope. Centralized repositories have limitations both in their failure tolerance and in their scalability. Existing peer to peer repositories either cannot guarantee to find query results even if these results exists in the network or require upfront definition. Hence the problem is to design a scalable indexing system.

New Ideas and Strengths:

Content- Addressable Networks proposed in the paper is the solution for a scalable indexing system. It uses hash tables that maps filenames to their location in the system. CAN is designed on d-dimension Cartesian coordinate space. CAN is useful for multi-attribute data. Efficient routing is a critical aspect of CAN. Node arrivals and departures affect a small number of existing nodes in a very small locality of coordinate space. Scalability can be achieved without increasing per node state. There are various other applications of CAN like large scale storage management system, name resolution services etc. While discussing the design of CAN the author discusses both the key problems i.e. scalable routing and indexing. The paper discusses various design improvements like increase in dimension, multiple independent coordinate spaces, overloading zones. The scalability of the design was validated by simulation results.

CAN is robust as it is completely distributed there is no single point of failure. It has properties like not exploring database recovery and resilience of routing. It supports low latency routing which is also validated by simulations.

Weaknesses and Extensions:

There are a couple of problems related to CAN like design of secure CAN which is resistant to denial of service attack. CAN algorithms should be extended to handle mutable content. CAN should support certain application specific, higher level primitives like keyword searching and anonymity. The paper does not discuss the effects of replication. Many of the optimizations involve replication of (K,V) pairs. Various under reasonable assumptions are made about replication limit and its effect on design parameters.

Download