روشی برای تولید مدل های شبیه سازی با کاربرد پارادایم تعامل وب

Modern internet and web applications rely on interactions among remote host computers connected by heterogeneous networks (different LANs, gateways, WANs, MANs, etc.). Simulation modelling such networks is of great importance to the web application designer to predict, at design time, performance metrics such as the end-to-end delay between hosts, which is dramatically increased by the various mechanisms necessary to deal with heterogeneity (protocol conversion, packet fragmentation and re-assembly, flow control, etc.). On the other hand, producing a simulation model of web interactions is a non-trivial task because of the great importance of the software aspects. It is thus necessary to provide general model production guidelines which can be then tailored and applied to specific simulation languages or packages. This paper gives such general production guidelines with an example application to the production of simulation models for web interaction paradigms of client–server and mobile agent types. An example use of the models is also introduced to predict the most convenient paradigm and the best choice of the host capacities for each given network configuration.

The performance of internet and web applications is a function of three factors: (f0) the performance of the interacting hosts; (f1) the performance of the heterogeneous network (LANs, gateways, WANs, MANs, etc.) that connects the interacting hosts; (f2) the interaction paradigms used to carry on the host interaction. Example paradigms are the Client–Server (CS) paradigm, the Mobile Agent (MA) paradigm, the Code On Demand (COD) paradigm, the Remote Evaluation (REV) paradigm, etc. Factor f1 is largely responsible of the end-to-end delay between hosts, because of the various mechanisms necessary to deal with heterogeneity (protocol conversion, packet fragmentation and re-assembly, flow control, etc.). Factor f2 also affects performance to a great extent since, depending on the interaction paradigms, the application load is differently distributed on the network, or on one or the other interacting host. According to [4], interaction paradigms can be characterized by: (i) the location of the software components before and after the execution of a given service; (ii) the computation of components responsible for the execution of code; (iii) the location where the computation of service takes place. This is better illustrated in Fig. 1 and Fig. 2 that consider a scenario in which a computational component A, located at host HA, needs the result of a service, and there exists another host HB that can be involved in the execution of the service.

A method to produce simulation models has been introduced. The method provides general model production guidelines which can be then tailored and applied to specific simulation languages or packages. In the paper the method has been applied to the production of simulation models of two web interaction paradigms (client/server—CS and mobile agent—MA) by use of the Modline/Arch package. An example use of the models has also been introduced to assist the web application designer to predict, at design time, the performance of alternative web interaction paradigms. Applications that can benefit of the method are distributed information retrieval, industrial process control, network management, network-based cooperative work and electronic commerce.