Abstract
The project involves simulating the behavior of an N-tier configuration of servers under various parameter settings using the discrete-event simulation language GPSS. The N-tier system consists of three tiers of servers namely web tier, application tier and database tier. Requests generated hit the web tier first and get routed appropriately to various tiers depending upon the type of the request.
The project simulates, and studies the behavior of an N tier system when parameters such as request inter-arrival mean, service time at each of the tiers, and network delay time are varied. The responsiveness of the system and its behavior are compared using the reports generated by GPSS. PLUS procedures in GPSS are used to intuitively represent the behavior of the system under different parameter settings. The model itself is modified. For example, the N-tier system is modified so that all arriving traffic passes through all the tiers till they get serviced irrespective of the type of request. A database request will pass through web tier, application tier and then onto the database tier, however the time spent at the web tier and application tier is neglected i.e. zero. An application type request will pass through the web tier, and then to the application tier and stop. In this case the time spent by the request at the web tier is neglected. The above two modifications of the model will provide useful insights into the behavior of the N tier system under various settings.
It has been seen that the base simulation model emulates the behavior of the physical system. One can easily relate to how the physical system would behave under different parameter settings by referring to the results produced by the simulation model for similar settings. Validity of Little's law is verified for the system. Utilization in terms of time as well as system resources and throughputs is monitored using the simulation model. A number of simulation runs are made to establish the significance of the variations in the experiment factors. F distribution testing in ANOVA is used to validate this significance. Hence the objective of the project is to provide means by which one can predict the behavior an N tier system without having to actually build a physical system. This in effect saves cost, resources and time, and exemplifies a major benefit of simulation techniques.