Research On Component Performance Modelling And Performance Management For Banking System

To understand performance model of application software can help to optimizesystem architecture and implement scientific deployment. Performance model representthe essential identity of a system,the research on system performance model has attractedmore and more attention from researchers, IT company and huge commercialorganization. Business and management activities in commercial banks have deeply relyon computer system,they also have stronger desire to IT system performance modelingtechnology and performance improvement suggestion. Although a lot of research methodand idea have been introduced to analyze enterprise application,but because modelconsistent with application component,rationality of indicators design,usability ofanalysis method, these make it difficult to find a suitable modeling method for bankingsystem.First, the differentiated performance requirements of transaction requests in bankingsystem have been analyzed in this paper. Under the limitation of resource condition, toguarantee the service quality of urgent requests and consider the complexity to interrupttransaction processing,the queuing model of non-preemptive M/M/1 with 2-classworkloads was promoted. By analyzing the embedded Markov chain of leaving requests,the indicators of queue length of requests in a group of scenarios has been inferred. Toovercome the restriction of math based analysis method, a simulator was developed. Itcharacters with single thread to act as multiple service processes. By analyzing servicelog produced by simulator, it is easy to get the service indicators of two type requests,such as response time, queue length, and quantitative influence each other.Second, the requirement deploying more applications on a platform was discussed.Consider components which service processes can transit between multiple states, If itsstart procedure, end procedure and no start state were treated as independent vacation,then, based on discrete time vacation queuing model Geom/G/1, a lifecycle schedulepolicy for the component was promoted,by Markov analysis, waiting time and queuelength indicators have been inferred. A simulator for the component was developed, It can help to analyze unsteady and multiple service processes component. More indicators areeasy to calculate from the simulation log,including requests queue length, systemresponse time, ratio of resource free time etc.Third, by analyzing the demand of multiple applications shared a cluster platform,the non-balanced schedule policy in a cluster was promoted that aimed to control theusage method of resource when workload is lower. Based on discrete time workingvacation queue Geom/G/1(WV), the performance model was established for componentcluster that employ the new policy. Performance indicators, such as the number ofrequests in system, response time of request, effective working time of both nodes wereinferred. A simulator for the cluster's performance model was developed also, throughanalyzing simulation log, more indicators may calculate.After analyzing component performance models and studying a performancemanagement processes in a commercial bank. A new method for improving performancemanagement was proposed. Performance model is the core object in the method. Thecontents in the method include architecture analysis, performance testing andperformance tuning. By integrating these tasks involved in software lifecycle, it will makeapplication performance model more clarity.