Research On Distributed Parallelizability Analysis And Optimization Of Legacy Code

As an emerging business computing model,cloud computing has many advantages such as parallelism,scalability and high reliability.In this context,it has become a trend to migrate legacy systems to cloud platforms.In the migration process,in order to fully utilize the parallel advantages of cloud computing,it is necessary to refactor the legacy code according to the cloud computing programming model,and the premise of the refactoring is that the legacy code can be parallelized distributedly.In order to analyze the distributed parallelizability of legacy code,decision rules DPDR(Distributed Parallelizability Determining Rules)are proposed in this paper,and are refined into four types of distributed parallelizability determining features: data dependency,continuous dependency,non-homologous and randomness.A distributed parallelism analysis algorithm is designed based these features.In reality,not all legacy code can be parallelized distributedly.An optimization method is proposed to optimize the code that cannot be parallelized distributedly.The optimization method divides the above four types of distributed parallelizability determining features into strong and weak types.Then,the legacy code is divided into three categories according to its distributed parallelizability determining features: legacy code with no feature,legacy code with weak features and legacy code with strong features.A source file organization method is proposed to make the input file of the legacy code with no feature conform to the input format of Hadoop,and a method is proposed for reorganizing source files to achieve distributed parallel for legacy code with weak features,and an iterative grading method is designed to enable legacy code with strong features parallel in part to improve performance.Finally,the tool DPAO(Distributed Parallelizability Analyzer and Optimizer)is developed and used to carry out experimental verification.Experimental results show that the tool can effectively analyze the distributed parallelizability of legacy code,the source file reorganization method can effectively make the legacy code distributed parallel,and the iterative grading method can effectively achieve partial parallel for legacy code.