Paper #: [5.1 P2P] 6
Title: Search and Replication in Unstructured Peer-to-Peer Networks
by Qin Lv, etc.


PROBLEM

Currently, unstructured peer-to-peer networks do not have optimal
search techniques because they suffer from problems such as flooding
the network and having duplicate messages reach the same node.
Consequently, there is a need to improve so that the minimal number
of nodes may be contacted for the maximum number of query results.
There is a movement to move from the flood-based query model in
systems such as Gnutella to a random walk method.  However, random
walk is not perfect because it has issues such as the walks causing
a longer search delay.

NEW IDEAS AND STRENGTHS

Some new ideas from this paper included using multiple random walks
working simultaneously to find answers to queries and to utilize
more replication over the network.  This is different from the
flood-based query model of Gnutella and has the benefit of finding
query results faster than the traditional random walk strategy.  In
addition to increase the efficiency of the multiple random walkers,
nodes will have states so that the node may send different walkers
for the same query to different neighbors.

The paper does a good job of summing up important goals in peer-to-
peer networking.  For example, the adaptive termination of queries
instead of using a standard TTL across the network, minimization of
duplicated messages, and minimization of affected nodes during
queries are all things that designers should strive for when
creating a peer-to-peer network.

WEAKNESSES AND EXTENSIONS

While the paper states a nice claim for using multiple random walk
searches in order to find results, the paper did not give any
numbers for the specific number of walkers that should be used.  It
only referred to the number of walkers as "k".  I would have liked
to have seen some statistics on efficiency with varying "k" values.

In addition, the paper also proposed replicating objects along the
path of the requester.  This is a good concept.  However, this
section should have been developed more.  I would like to know if it
would take into consideration the resources of each node when
determining which nodes to use for replication if random replication
was used.