Many computer networks can be loosely classified as unstructured networks. These
networks include unstructured peer-to-peer networks, wireless ad hoc networks,
sensor networks, mesh networks, etc.
There is neither a central
directory server nor any control over the network topology or data
placement. The computing nodes create impromptu connections in an often loose and random fashion. The placement of data
is not based on any knowledge on the topology. The applications of these networks
are very broad, and they have attracted significant attentions from the industry and the
academic research.
The main advantage of unstructured networks is they do not require high maintenance
overhead. Indeed, many applications require the networks do not need any maintenance
at all. However, unstructured networks have their disadvantages and scalability issues.
Just to name a few of the critical research problems.
(1) Efficient search in unstructured networks is difficult.
Since there is neither a central directory server nor distributed indexing
service, the best a computing node can do is a blind search. For example,
flooding search causes excessive traffic while random walk based searching
requires long latency and is non-deterministic.
(2) Content and service replication is especially important but also difficult.
Since the computing nodes are less aware of each other, it is often a good idea to
replicate data (or some clues about the data) in the networks. This will facilitate search and
discovery, and will make the data more available to the nodes. However, since the networks can
be rather complex, it is difficult to find fully distributed, efficient replication schemes.
(3) Dynamics of the networks make everything challenging. For example in peer-to-peer
networks, the peers may join and leave the network frequently; in wireless ad hoc
networks, the nodes may move around; in sensor networks, the sensors may die out after energy
is consumed and new sensors are distributed again. It suggests that may
techniques, e.g., routing and search/discovery, should take into consideration such dynamics,
and if possible, even take advantages of such dynamics.
The UNO project aims at several objectives. (1) Design and performance evaluation
of search, information delivery, and replication techniques. We are particularly interested
in techniques that can be used in the general unstructured networks. (2) Analysis and optimization
of these techniques. We will study the limits of more conventional techniques and provide
insights on the performance of these techniques. We are particularly interested in the network
connectivity, search and replication schemes. (3) Exploration of novel network architectures
and protocols that can overcome the limitations of unstructured networks. We will look for
hierarchical and hybrid architectures, design novel information delivery and replication schemes, and
make the networks more robust, more scalable, and more efficient.
