Abstracts

There are certain classes of computing problems so complex or massive that they require the resources of more than just a single computer. These problems fit well in the mold of distributed processing. In the science world the problems are typically CPU intensive, whereas in the business world they tend to be more input-output intensive. Although there is some carryover of the basic logic, it is difficult to transfer whole parallel or distributed algorithms directly from science to business. [1]. Therefore, there is a need to undertake simulation and modeling projects specifically geared to business applications that because of performance concerns are candidates for distributed processing. In the business IT world response time to the end user is often a primary benchmark used to measure the effectiveness of a system. Past modeling research has used these criteria to measure performance of network architectures [2] and distributed web servers [3]. It therefore, seems logical to apply the same modeling techniques to distributed database models to determine baseline performance information.

In this study a business oriented WAN will be programmed through simulation and the following will be varied: number of node storing the database, number of workstations accessing the database, and the probability of any given request being sent to a given database node. The data collected in these simulations will be used to test the following null hypotheses:

H1. The number of nodes a database is stored upon has no effect on the delay to the end-user workstations.

H2. The number of end-user workstations making inquiries to a distributed database has no effect on the delay of the responses back to the initiating workstations.

H3. The probability distribution that controls which inquiry will be assigned to which database node has no effect on the delay of the responses back to the initiating workstations.

The software for the basic simulation has been written and the following data was collected. The delay for a 10 workstation network in which the database was stored on a single node was 207 milliseconds. When a second database node was added with a 0.5 probability that any request would be assigned to either database node, the delays were 195 milliseconds if assigned to node 1 and 160 milliseconds if assigned to node 2. These results provide evidence that a multiple node database design may help reduce delays, however, a significant number of trials under varying conditions will be needed to collect enough data to test the hypotheses stated earlier. The full paper will incorporate that data and the analysis thereof.

References

[1] Guster, D., Sultanov, R & Chen, Q. (2003). "Adaptation of a Parallel Processing Technique Used to Solve a Physics Problem to a Computer Network Management Application", Accepted for presentation and publication in the proceedings of the Information Resources Association International Conference, Philadelphia, PA.

[2] Guster, D., Sohn, C., Robinson, D. & Safonov, P. (2003). A Comparison of Asynchronous Transfer Mode (ATM) and High Speed Ethernet and the Network Design Implications to a Business Organization, Journal of Information Technology and Decision Making. In Press.

[3] Guster, D. et al. (2003). Using Simulation to Predict Performance Characteristics of Mirrored Hosts Used to Support WWW Applications. Issues in Information Systems. In Press.

Full Text in Russian:

ZIP (108 kb)

Note. Abstracts are published in author's edition

Comments

[Home] [Conference]

© 1996-2000, Institute of computational Techologies SB RAS, Novosibirsk
© 1996-2000, Siberian Branch of Russian Academy of Science, Novosibirsk