Paper #: 5.1.13
Title:  A Scalable Content Addressable Network

1. Problems
-distributed hash table-like infrastructure on an Internet-like scale
-scalable, fault tolerant, self-organizing

2. New Idea and Strengths
-Content Addressable Network (CAN) that lays out a structured file system that with 100% accuracy on queries. Because of the hashing function, each key (the search value) maps to the location of the file. Retrieving the file whose location is known consists of traversing the structured network.
-The network construction and mainenance is sound, with good upkeep amongst neighbors and a good division of bandwith according to neighbors' responses.
-Exremely scalable, as increases in nodes barely affected overall response times, while maintaing the stability of the smaller systems.

3. Weaknesses and Extensions
-A glaring hole in the explanation is how files end up at the nodes they are supposed to be stored by. Nodes randomly pick a space to occupy in the network, yet they are also responsible for delivering the specific files that hash to their space. I tried to find any account of this, but it appears they never mention it. They do not discuss the hashing function either, or any ideas regarding keyword searching (although they do mention the last fact).
-Although the system appears scalable, they did not test large numbers of nodes, beyond 230k. Hard to tell if a problem is looming with larger numbers of nodes.
-Did not discuss how files would be transfered- directly once the node was found, or transfered back along the nodes.
-The "testing" appears to be assumptions using chosen numbers for latency, which hardley seems realistic. Little justification is given for the choice of numbers.
-They fail to successfully take advantage of geographical proximity. Nodes naturally gravitate towards nearer neighbors, as they tend to provide quicker responses, but once a node joins the network it does not move, it merely adjusts its choice of neighbors.
-Possible Extension: Make insertions less random. In order to minimize the resulting grouping, assign dense groups of nodes a large area to subdivide amongt themselves, in order to maximize the benefit of close neighbors.