As broadband connections to the Internet become more common, new
information sharing applications that provide rich services to
distributed users will emerge. Furthermore, as computing devices
become pervasive and better connected, the scalability requirements
for Internet-based services are also increasing. Distributed object
middleware has been widely used to develop such applications since
it made it easier to rapidly develop distributed applications for
heterogeneous computing and communication systems. As the
application's scale increases, however, the client/server
architecture limits the performance due to the bottleneck at the
centralized servers. The recent development in peer-to-peer
technologies creates a new opportunity for addressing scalability
and performance problems for services that are used by many nodes.
In a peer-to-peer system, peer nodes can contribute a fraction of
their resources to the system, enabling more flexible and extended
sharing between the entities in the system. When peer nodes are
required to contribute their resources by replicating a service for
self and others, however, several new challenges arise.
Our thesis is that non-dedicated resources in a distributed system
can be utilized to replicate shared objects dynamically so that the
quality and scalability of a distributed service can be achieved
with lower cost by replicating the objects at right places and
updates to those shared objects can be disseminated efficiently and
quickly. The following are the contributions of our work that has
been done to validate the thesis.
1. A new fair and self-managing replication algorithm that
allows distributed non-dedicated resources to be used to improve
service performance with lower cost.
2. A multicast grouping algorithm that is used to disseminate
updates to the shared objects among a large set of heterogeneous
peer nodes to keep consistent view for all peer nodes. It groups
nodes with similar interests into same group and multicasts all the
required data to the group so that the unwanted data received by
each node can be minimized.
3. An overlay construction algorithm that aims at reducing both
network latency and total network traffic when delivering data
through the built overlay network.
4. An implementation of a distributed object framework, GT-RMI,
that allows peer nodes to invoke dynamically replicated objects
transparently. The framework can be configured for a particular peer
node through a policy file.
| Identifer | oai:union.ndltd.org:GATECH/oai:smartech.gatech.edu:1853/7589 |
| Date | 29 November 2005 |
| Creators | Chang, Tianying |
| Publisher | Georgia Institute of Technology |
| Source Sets | Georgia Tech Electronic Thesis and Dissertation Archive |
| Language | en_US |
| Detected Language | English |
| Type | Dissertation |
| Format | 795471 bytes, application/pdf |
Page generated in 0.1443 seconds