2012 May 23 at 09:00
DC 2310
Kimiisa Oshikoji, graduate student, David R. Cheriton School of Comp. Sci., Univ. Waterloo
This talk describes a methodology by which sub-performing library versions can be identified from the behaviour observed from different library configurations. These are measured by time, static, and dynamic analyses while executing the test suite of a project. During the course of these runs, different configurations of the libraries are substituted in and the results collected to be analyzed.
This topic is important because there are many different versions of a library for a developer to choose from when building a software system. Once the library is chosen, the developer must determine which version of the library to use. Is there some characteristic that can identify the correct version of a library to use? Even if a library compiles correctly, there could be a better version of that library that will provide superior performance. In particular, the developer would prefer to avoid poor configurations: that is sets of libraries that perform poorly, or not at all.
The results of this analysis shows that there is no quick way to identify a sub performing library; however, sub-performing libraries can be identified through concentrated efforts to collect and analyze time-based data enabling developers to avoid sub-optimal system configurations.