Analyse et optimisation de patterns de code

Our work is performed in the context of a financial company that develops and maintains legacy software. This software has been existing for more than twenty years, it was written in a proprietary, procedural and 4GL language called ADL (or Application Development Language). This software was initially developed for VMS system and deals with old generation of DBMS. For commercial reasons, the software has been ported to UNIX systems and to new RDBMS; Oracle and Sybase. It has also been extended to offer web interfaces. This legacy software has to face some new challenges as databases grow. During these last 20 years, some phenomenons like the merging of two companies, make data grow up to more than 1Terabyte and will reach 1Petabyte in a few years. In these new contexts, the ADL language shows limits to handle such a mass of data. Some patterns of code with database access have been suspected to be responsible for important decrease in performance. Our work consists in detecting all the instances of a suspected pattern of code in the source code or procedures, and identifying the instances of code, the most often called and the most time consuming. We developed a first tool called Adlmap, which is based on static analysis. It detects all DB accesses and flags those that are suspected patterns of code. The second tool we developed, named Pmonitor, is based on hybrid analysis; a combination of static and dynamic analysis. We use it to highlight inefficient code patterns instances.