Preliminary Research Proposal
Resilient and Robust Distributed System
University of Rochester
Department of Computer Science
PO Box 270226, Rochester, NY 14627, USA
Tel: (585)275-5671, Fax:(585)273-4546
The 21st of September, 2005
Introduction
The Internet is the largest distributed system infrastructure. From the very beginning the dependency on Internet has increased day by day. Now the scenario is much different from the beginning. Various security and service critical applications are run over internet. Various government organizations, hospitals, banks, insurance companies, service organizations and many other crucial institutions extensively use Internet for their activities. Each of these organizations operates over a distributed system. As these distributed systems are running over Internet those are vulnerable to various adverse conditions. So to keep these institutions operational on every adverse situation the distributed system should be robust, resilient and secured all the time.
With the help of Computer and Information Science and Engineering (CISE) of National Science Foundation (NSF) we the researchers of systems group of the Department of Computer Science of the University of Rochester has set target to develop a more secured, robust and resilient distributed Infrastructure.
Motivation
With the time the speed and storage of individual node have increased rapidly. Now even a personal desktop computer has huge computing power and every one of those can work as a server itself. But these huge resources are kept almost idle. So to utilize these resources a distributed system can be developed. This utilization will produce a distributed system having loosely cooperating but administratively unrelated nodes.
Currently the distributed systems are built from two different models. One is client server model and another is overlay model. In client server model a single server or localized cluster of servers are deployed and all clients request to the(these) server(s) and server responds as usual with the data requested. E-mail, search engines, e-governess, distributed file systems, electronic shops etc are the examples of centralized client server model. In overlay model many servers located in different locations connected through Internet form an application-level and application-specific intranet situated on top of Internet. FFnet is one such example.
Both centralized client server model and overlay model is controlled by centralized administrator although in overlay model the servers are decentralized. Both have distributed computing but both of the models are subjected to single point of failure. Moreover their performance degrades substantially with failure of one of the nodes.
Objectives
So our objectives are
To utilize the unused resources at PC level.
To design a more Secured Distributed System that can works being inside untrusted nodes.
To make the system really distributed that is decentralized administration.
To make the system more Robust under failures, attacks and high loads.
To make the system highly Resilient against attacks and failures.
To make the system more scalable to integrate huge number of resources.
Methodology
Use Dynamic Hash Table (DHT) based content retrieval approach.
Development of large-scale decentralized application.
Use modified overlay model to integrate the distributed PC level (some of those may be untrusted) resources.
Incorporate dynamic administration, load balancing system so that the system is not subjected to single point of failure.
Qualifications of the Team
The Department of Computer Science of the University of Rochester has unique reputation in the research in the systems area. Although the department is a small one but it has some outstanding faculty members who have proved their excellent level of research many many times. They have contributed to the CISE and also doing contribution now. Professor Michael Scott, Professor Sandhya Dwarkadas, Professor Chen Ding and Professor Kai Shen are among those in the systems group. Their research are supported by NSF grants including NSF Career awards.
This system group has long experience to do research in parallel and distributed systems, scalable Internet overlay network, i/o intensive servers, multi-component network services and middleware system. This research proposal will be an extension to the the ongoing researches and past researches reflected through their publications' list.
Resource Requirement
We shall require around 500 PC for initial testing. We can request some PC manufacturer like IBM or Intel to have to use such number of PCs. The estimated fund required in this project is 1 million dollar.
Conclusion
So undoubtedly we can infer that by providing the funding support in this project the NSF-CISE can take the current state of the Internet based Distributed System to a new superior level. As a result the boom in Internet based services will continue its expansion with a more increased rate.
References
Moore, D., Voelker, G. M., and Savage, S. Inferring internet denial-of-service activity. In Usenix 2001 security conference (2001).
Adjie-Winoto, W., Schwartz, E. and Balakrishnan, H. and Lilley, J. The design and implementation of an intentional naming system. In Proc. ACM Symposium on Operating Systems Principles (Kiawah Island, SC, Dec. 1999), pp. 186-201.
Andersen, D., Balakrishnan, H., Kaashoek, M. F., and Morris, R. Resilient overlay networks. In Proceedings of the 18th ACM Symposium on Operating Systems Principles (SOSP '01) (Chateau Lake Louise, Banff, Canada, October 2001