Agile Business Suite Log Analysis and Stochastic Modeling of Transactions

Abstract

Agile Business Suite (ABSuite) is an application development and deployment product that can define, generate and manage complete, highly scalable, real world, platform independent applications. The product was developed by Unisys Corporation with the intention to allow developers to rapidly develop their complete applications without having to think about low level platform dependent implementation details thus saving them from having to write thousands of lines of code. Many enterprises like Banks, Healthcare Facilities, Traffic Management, Financial Institutions and etc. use ABSuite to develop their applications and automate their business logic. Once deployed on the application servers, the generated applications are managed by the ABSuite Runtime framework. One important function of the ABSuite Runtime is to maintain a set of log files that contain immense information on system behavior, communication between the system and the runtime framework and user interaction with the system. There are various kinds of log files that are generated by the ABSuite Runtime, each having its own format and vast amounts of hidden knowledge stored in them. These log files contain important information regarding usage patterns, system failure patterns and other performance bottlenecks. Thus proper analysis of these log files is necessary to obtain vital information which will help in optimizing system performance, easing maintenance tasks, identifying hidden bugs and abnormal behavior and adopting better design strategies. Log files are produced by almost all devices, systems and protocols. In general, the analysis of any kind of log file is not an easy task. There are several challenges that need to be addressed first in order to obtain proper and accurate results. The first issue is directly related to the fact that ABSuite log files are often huge and heterogeneous in nature. Standard Algorithms fail in such conditions and thus the task of log analysis requires a different approach. Secondly, identification of required information and the log files storing such information requires domain specific knowledge and expertise regarding the generated log files. Without prior information on what to look for and where to look for, the analysis process will become impossible. Last but not least is the issue of unstandardized format of each of the ABSuite log files which makes processing a difficult task. Keeping the above things in mind, this thesis presents some of the ideas behind developing a tool that automatically analyzes the generated ABSuite log files and extracts information that will help the developers to optimize the application’s performance and reduce the system’s downtime and maintenance cost. vii In this thesis we discuss three stages of analysis. The first stage (Ispec Analysis) finds basic statistical parameters like trigger count, frequency and probability distribution of each transaction occurring in the system. This information is used in subsequent stages to obtain more accurate results and hypothesis. In addition, We also present a way to use the results of this stage to estimate usage and traffic patterns, peak loading conditions and heavily used modules of the system. The second stage (Response Time Analysis) obtains mean time to response for each transaction. This information is used to identify transactions with high latencies and find the most likely cause of those latencies. The third stage (Exception Analysis) finds which transactions frequently generate exceptions and how many types of exceptions are generated by each transaction. The results of this stage help in identifying the number, type, severity and cause of exceptions generated by the system. The results of this stage can also help in taking high level decisions like whether system resources need to be increased to avoid deadlocks, whether there is any need to increase bandwidth allocation, whether there is any need to change the current design of the system and etc. Finally, we present and compare the analysis results of all the three stages for two sample applications (developed using ABSuite) with known usage characteristics. We also explain how the developers can use the results of each stage of analysis to better optimize their systems using the two sample applications as case study.