Reliable Data Sharing in Wide Area Gigabit Networked Databases

Panos K. Chrysanthis                                        Sujata Banerjee
                    Dept.  of Computer Science                   Dept. of Info. Sci & Telecommunication
      University of Pittsburgh                                   University of Pittsburgh

Contact Information

Panos K. Chrysanthis                                   Sujata Banerjee
Dept. of Computer Science                          Dept. of Infor. Sci. & Telecommunications
220 Alumni Hall                                           749 SIS Building
Pittsburgh, PA 15260                                   Pittsburgh, PA 15260
Phone: (412) 624-8924                                Phone: (412) 624-9470
Fax :    (412) 624-8854                                Fax:     (412) 624-2788
Email: panos@cs.pitt.edu                            Email: sujata@tele.pitt.edu
URL: http://www.cs.pitt.edu/~panos         URL: http://www.pitt.edu/~sujata

WWW PAGE
 
URL:  http://www.cs.pitt.edu/admt/Projects/distributed_databases.html

List of Supported Students

• George Kyrou  and Anand Deshpande,  Graduate Student Researchers

Project Award Information

• Award Number: IIS-9812532
• Duration: 09/01/1998 - 08/30/2001
• Title: Reliable Data Sharing in Wide Area Gigabit Networked Databases

Keywords

Data servers, Caching, Transaction Processing, High speed networks, Data consistency and recovery

Project Summary

This project's goal is to study the impact of the new capabilities of high-speed networks on the design and performance of data management protocols. In particular, the focus is on high-speed wide-area networks (WANs) where the propagation latency is the dominant communication cost, the migration of large amounts of data is not an issue, and efficient multicasting and guarantees on network Quality of Service (QoS) parameters are available. In this project, existing concurrency control, commit and recovery protocols are being evaluated with respect to performance scalability under the new network assumptions. Further, new protocols are being developed that utilize the huge bandwidth and minimize the number of sequential phases of messages exploiting the multicasting capability and the provision of QoS guarantees such as bounded end-to-end packet delay and packet loss. At the same time, because of the need to scale to thousands of clients and servers, not all of which may be operational at any given time, the new WAN data management protocols are designed to achieve higher site autonomy in a cooperative manner, reducing global synchronization and supporting disconnected operations. The various protocols are empirically evaluated using computer simulation under different scenarios. The results of this research are expected to provide a better insight into the synergy between database and network technologies. This, in turn, will facilitate the deployment of future high performance WAN data server systems which are part of the necessary infrastructure that will support important nation- and world-wide applications such as electronic commerce.

Goals, Objectives, and Targeted Activities

This year our goals and objectives were as given below.

• Development and evaluation of two read-only optimizations for the g-2PL protocol with the goal of obtaining even better system scalability.
• Enhancement of the simulation testbed to incorporate multiple servers and various transaction workloads.
• Continue the evaluation of the scalability of commit protocols for multi-level transactions in a WAN.
• Development of methods to ensure the consistency and currency of client read-only transactions in the presence of updates in data broadcast environments and study the important case of multi-server broadcast environments.
• Study and development of future database applications as a direct result of advances in networking capabiliies.

Indication of Success

We have made substantial progress on the objectives stated above. Work on each of the different aspects of the project are progressing well, and we are in the process of analysing the simulation results. We have made significant progress on formulating the multi-server problem in the broadcast environments as well as developing algorithms to maintain data currency in these environments. Two new database applications are being considered as part of this project, that are motivated by continuing changes in the underlying networking infrastructure. The first application falls within the purview of E-commerce, and deals with the dynamic formation and dissolution of virtual enterprises based on dynamic workflow management systems. An important component of this work has been to collaborate with the AI planning community with a view to re-using the relevant results from dynamic plan management. The second application is that of distributed mobile GIS systems in which emergency and disaster recovery operations are targeted. A new 3-tiered architecture has been developed as part of this research.

Project Impact

• One student completed his MS and two more are expected to finish their MS degrees this academic year. Undergraduate students will be involved through the REU program.
• Based on the research conducted in this project, the course contents of two Ph.D. level courses on data management are being augmented.

GPRA Outcome Goals

The goals of this project are in consensus with the long-term vision of the computer and communications industry to provide fast and reliable information access to the masses, taking into account the individual user's specific needs. Every attempt is being made to develop theoretical frameworks and then transfer them into pragmatic environments so their impact is far-reaching in the commercial sector. The project thus far has attracted a diverse body of students, including women, who are particularly under-represented in this research community.

Project References

These and other papers are available from the project WWW page.

• S. Banerjee and P. K. Chrysanthis. Data Sharing and Recovery in Gigabit-Networked Databases, Fourth International Conference on Computer Communications and Networks, pages 204--211, September 1995.
• S. Banerjee and P. K. Chrysanthis. Network Latency Optimizations in Distributed Database Systems, International Conference on Data Engineering, pages 532--540, February 1998.
• C. Khunboa and S. Banerjee. Network Traffic Characterization of Distributed Database Applications. Annual Simulation Symposium, pages 98--105, April 1998.
• S. Banerjee and P. K. Chrysanthis. Group Two-Phase Locking: A Scalable Data Sharing Protocol, IEICE Transactions on Information Systems, Vol. E82-D, No. 1, pages 236—245, January 1999.
• P. K. Chrysanthis, T. F. Znati, S. Banerjee and S. K. Chang, Establishing Virtual Enterprises by means of Mobile Agents, Research Issues in Data Engineering Workshop, March 1999.
• Y. J. Al-Houmaily and P. K. Chrysanthis, Atomicity with Incompatible Presumptions, Proc. of the 8th ACM SIGMOD-SIGACT-SIGART Symposium on Principles of Database Systems, pages 306—315, June 1999.
• E. Pitoura and P. K. Chrysanthis, Exploiting Versions for Handling Updates in Broadcast Disks, Proc. of the 25th Int'l Conference on Very Large Data Bases, pages 114—125, September 1999.

Area Background

The same reasons that caused the emergence of the client-server database systems to become the predominant distributed database architecture in local area networks (LANs), combined with ever faster and cheaper computers, cheaper stable memory, high-speed networks and network-aware applications, are now causing unprecedented growth in information/distributed database systems over WANs. The WWW is a prime example of a WAN data-server system which is primarily read-only for the time being. In the future, it is expected that general data-server systems (also called data shipping or enhanced client-server systems) in which clients perform much of their query and transaction processing locally, will be deployed over wide-area networks (WANs). It is also expected that multiple WAN servers will operate cooperatively and that WAN clients will evolve and function as data-servers over local-area networks that will also need to cooperatively process transactions across multiple WAN servers, creating an even more powerful distributed database environment of multiple data-servers and collaborating-servers over WANs. It is within such a database environment that we envision that nation- and world-wide enterprises, for example, will be operating and electronic commerce will be carried out. In order to realize such powerful distributed database environments, there is a need to study the impact of the new capabilities of high speed networks on the design, the performance and the scalability of data management protocols, specifically the availability of huge bandwidth, of efficient multicasting and of Quality of Service (QoS) guarantees.

Area References

• K. Wilkinson and M. A. Neimat. Maintaining Consistency of Client-Cached data, International Conference on Very Large Databases, pp. 122--134, August 1990.
• Y. Wang and L. Rowe. Cache Consistency and Concurrency Control in a Client/server DBMS Architecture, ACM SIGMOD International Conference on Management of Data, pp. 367--376, May 1991.
• L. Kleinrock. The Latency/Bandwidth Tradeoff in Gigabit Networks, IEEE Communications Magazine, vol. 30, pp. 36--40, April 1992.
• A. Delis and N. Roussopoulos. Management of Updates in the Enhanced Client-Server DBMS, International Conference on Distributed Computing Systems, June 1994.
• M. J. Franklin, M. Carey, and M. Linvy. Transactional client-server cache consistency: Alternatives and performance, ACM Transactions on Database Systems, 1997.
• S. Venkataraman, J. F. Naughton, and M. Livny. Remote Load-Sensitive Caching for Multi-Server Database Systems, International Conference on Data Engineering, February 1998.

• Crafting Recovery for Advanced Transaction Models and Applications: Mobile transactions are instances of advanced transaction models.
• Transaction Management in Multidatabase Systems: One can view mobile data management system as a special case of a multidatabase system. For example, the notion of local autonomy in mobile environments is manifested in the ability of mobile computers to continue to operate in an independent fashion when disconnected.
• Distributed Constraint Maintenance: Efficient maintenance of integrity constraints that will require little, if any, coordination between mobile computers and database servers, will both facilitate disconnected operations as well as reduce the cost of communication during periods of adequate network connection.
  Mobile computers and wireless networks are becoming an integral part of distributed computing environments which means that there is great potential for collaboration with projects focusing on wide-area database systems and heterogeneous and semi-structured data management.