PopCon: Mining Popular Software Configurations from Community

Rukma Talwadker, Deepti Aggarwal; NetApp Inc

2019 ACM/IEEE International Symposium on Empirical Software Engineering and Measurement (ESEM)
September 16-20, 2019
PORTO DE GALINHAS, BRAZIL

Software system configuration problems are fairly prevalent and continue to impair the reliability of the underlying system software. Configurations also play an important role in establishing the quality of the software. With every configuration “knob” we delegate a responsibility to the user and also, we might make the software vulnerable to a failure, poor performance and other system operational issues. Efforts to facilitate a healthy configuration can be summarized by the way of following steps: 1) Gain knowledge about what defines a configuration; 2) operationalize a mechanism to mine popular or recommended configuration defaults; and 3) leverage insights for improving software quality or faster troubleshooting and fixing in the case of a software failure. Using PopCon, a tool that we built, we target all three aspects in a closed-loop fashion, by focussing on storage system software from NetApp, ONTAP data management software. We learn popular configurations from the deployed community, evaluate active configurations, deliver actionable information through this tool. Our findings have been encouraging. We can report that about 99% of our ONTAP software user community gravitates towards popular configuration values. Though about 20% of the configuration parameters initially need a custom or user input, we have found that over a period of a few months, systems adopt these popular values. Also, there is a high correlation between the number of outstanding deviations from the popular values and the number of active support cases on these systems. Further, we have also learned that for about 40% of the systems with support cases, deviations disappear at about the time of case closures. Finally, PopCon capabilities presented here are simple to implement and operationalize in any software system.

Resources

  • A copy of the paper can be downloaded at: link.