Grant: Discovery Grants Program - Individual

Computer scientists are often encouraged to make their research reproducible so that other scientists can verify, reproduce, and extend their experiments. However, reproducibility in computational sciences is a goal hard to achieve due to the increasing complexity of computational experiments and implicit dependencies on data and execution environment usually involving many steps and combining several data sources. In order to support reproducibility and to deal with the increasing complexities of big data experiments, the use of scientific workflow management systems (SWfMSs) to explore data, plan experimental execution, and visualize results is becoming more common. Given that diverse varieties of software data are generated throughout the software lifecycle, exploiting the scientific workflow-based approach for software clone analytics could be a potential way to achieve large scale data intensive discoveries that would not only support reproducibility but also would bring the other benefits of such systems. However, adapting an existing SWfMS to a new domain such as software analytics is challenging since a reasonable number of domain specific tools must need to be available and to be adapted to be used in workflow systems to be sufficient for composing different workflows for analytics. Furthermore, usability and flexibility for scientists to incorporate their own domain knowledge in the workflow systems are also pressing challenges for adapting to a new domain. I will address this aspect for software clone domain by integrating a good number of tools into the proposed workflow framework and providing reusable workflows. In the short term, (1) I plan to build a cloud-based SWfMS (called SciClone) which will not only provide a reproducible environment for software clone analytics but also will exploit the features of a traditional workflow management system in conducting analytics tasks in the cloud by composing workflows using drag and drop graphical elements and/or simple domain specific scripts, and without worrying about computational issues or other technical details, and (2) I will conduct ground-breaking scientific discoveries in software clone analytics in this unified framework with large scale datasets and hundreds of different possibilities which were never possible before. The long-term objective is to build a fully-fledged SciClone, and not only for software clone analytics but also for general software analytics where other researchers could also conduct their scientific discoveries by the already available tools and workflows in SciClone or possibly integrating a few required ones at ease. This proposal will train 3 PhD, 3 MSc and 6 UG students and will drastically change how reproducibility issue could be addressed and how software analytics should be performed, especially for large scale analytics of diverse varieties of software data in a unified framework exploiting the benefits of a scientific workflow-based approach.