Performance Profiling And Evaluation Of A Large System Based On PEPA

Performance Evaluation is used to describe, analyze and optimize the dynamic behavior of systems, which means it aims to analyze the performance of systems quantitatively. The traditional Performance Evaluation methods mainly include the Queue Theory and the Stochastic Petri-Net. With the popularity of personal computers and the spread of networks, the research on the Performance Evaluation of the large systems is becoming an urgent and important problem. However, the state space explosion problem makes the traditional Performance Evaluation methods computational infeasible in large systems. In order to solve this problem, this thesis chooses PEPA as the language to build the performance models of systems and provides a solution to the Performance Evaluation of the large system using fluid analysis algorithm. The main contributions of this thesis are as follows:1. The principle and implementation of the main Performance Evaluation methods are introduced. Then, these common Performance Evaluation methods are compared and PEPA is chosen to be the language used in this thesis.2. The syntax of PEPA is introduced and the techniques used to simplify the model are provided using the extension format of PEPA, which is GPEPA. Finally, in order to completely solve the state space explosion problem, fluid analysis, which solves the performance analysis problems by solving the ordinary differential equations which describe changes of the number of fluid components in a particular state of GPEPA model, is described and implemented in this thesis.3. Experiments are designed to obtain parameters of the performance models. This part mainly includes the implementation of experiments, the amendment to the datasets from experiments and the distributions fitting from the datasets. Specifically, the experiments include measuring the speed of the read/write and the file transform by dd and scp commands on Linux separately; the amendment to the datasets depends on the box-and-whisker plot and the distributions are obtained by Hyper Star using the cluster algorithm.4. Based on the parameters obtained from the experiments, GPEPA is used to model and simulate the performance of systems. Then, the passage-time and throughput of the system are evaluated to test the influence of the remaining parameters in the model. Also, a special component in PEPA, which is called “probe” component, is introduced to measure the passage-time of both the single action and the sequences of actions. Finally, using the data from the evaluation, some empirical conclusions are explained quantitatively and some advice about the optimal deployment of large systems are provided.