Performance Analysis Of Distributed Real-Time Systems In Grid Environment With Aspect-Oriented Technology

Performance is an important indicator of software's quality. With the rapid development of grid computing technology, more and more grid-based applications are developed and put into use. The heterogeneous, dynamic and self-governing resources make distributed real-time system in grid environment one of the most complex software nowadays. Its performance is particularly critical.Usually the performance evaluation is placed after forming of product, but that causes a lot of manpower and financial remedies when the problems emerge. It's obviously difficult to make up for such cost by "fix-it-later" approach, because actually most of significant performance issues are due to the basic architecture or design factors. Software Performance Engineering advises that study on the software architecture should be adopted in the early stage of software development to predict performance and identify bottlenecks in advance, making it easier and less expensive to take appropriate design decisions to solve performance problems.Aspect-oriented technique deals well with the modularity of crosscutting concerns. One of its core ideas is that different aspects can be described and modeled separately with different formal methods. The time analysis of distributed real-time systems is very complex and different functions may have distinct demand on time property. According to the AO technique, the time aspect could be decomposed into three sub-aspects, which are deterministic time sub-aspect, undeterministic time sub-aspect and fuzzy time sub-aspect, to facilitate the use of appropriate formal language.This paper proposes an aspect-oriented performance analysis process, which conducts performance modeling and evaluation early in software development. The key point is that performance scenarios involving different time sub-aspects should be separately modeled and analyzed using the most suitable techniques. The analysis results of different aspects are integrated as an important basis for performance evaluation. In this process, firstly we identify scenarios concerning different time sub-aspect; then according to SPE process, attach SPT performance annotations to UML design views, including structure diagrams and behavior diagrams, to express performance parameters in a normalized form; after further transformation from UML designs to formal methods, detailed analysis results will be obtained through corresponding automatic tools.A distributed web application paradigm is illustrated in detail for introducing an approach of transformation from UML into LQN performance model.