چكيده به لاتين
Todays, the power and performance management of computing systems is an important issue. On one hand, system providers are interested in designing and providing systems that use less power; on the other hand, users usually wish to have a high-performance system. Since, two measures, power consumption and performance, are inversely proportional, they should be simultaneously investigated. Different factors such as physical components, communication networks, cyber and software activities, and environmental conditions can affect both power consumption and performance of systems. Since the comprehensive evaluation of power consumption and performance of real-world computing systems is not feasible in terms of budget and time limitations, the modeling techniques can be used for this purpose. Therefore, it is an important research effort to propose an analytical model to investigate the power consumption of the system and evaluate the performance, at the same time.
In this thesis, in the first step, the important factors that affect power and performance measures of computing systems are identified and categorized. Then, applying extensions of stochastic Petri nets, different modeling patterns are proposed to evaluate the impacts of power versus performance measures in some case studies. Since stochastic Petri nets and their extensions are basically introduced for performance and dependability evaluation of computing systems, they do not provide appropriate facilities for direct modeling of power sources and evaluation of the power consumption. The results of case studies revealed that there is not a simple approach to specify the power sources, one of the most important elements of computing systems, in these formalisms.
In order to fulfill this requirement, in the second step, a new extension of coloured stochastic activity networks, called PCSAN, is proposed. This formalism, in addition to support the modeling patters proposed in the first step, provides a new primitive, called “Source”, to specify both permanent and rechargeable power sources. The PCSAN models can be transformed into the models described by existing formalisms and then can be solved or simulated using their analytical solvers or simulation techniques and tools. Finally, we present the results of a case study to illustrate how the power and performance of computing systems can be modeled and evaluated using the PCSAN formalism.
