ISSRE 2006 START Conference Manager    

Improving Test Effectivenss by Analyzing Customer Usage

Theresa Kratschmer, Hema Srikanth and Randy Adair

The 17th IEEE International Symposium on Software Reliability Engineering (ISSRE 2006) -- Industry Practices (ISSRE 2006)
Raleigh, North Carolina, USA, 6-10 November 2006


Abstract

Orthogonal defect classification (ODC) has been shown to be very effective for classification of in house defects and field failures. In the past, defect classification via ODC has been shown to minimize field failures by increasing test effectiveness. ODC has been applied to identify focus-areas of test to meet a desired level of test criteria before releasing the product. We apply ODC to determine customer usage by analyzing field failures of several releases of drivers written in Java. Our results indicate that majority of field failures could have been minimized by increasing basic testing of single functions.

We analyze field failures by examining the trigger, activity and impact attributes of ODC for five releases of driver APIs written in Java. Our goal was to classify the failures and apply the results to improve test in subsequent releases. Trigger, that indicates customer usage, was the first attribute to be classified. Activity is identified after classifying trigger based on trigger to activity mapping. Our results from classifying for activity show that 67% of the escapes were function test; 24% from system test; 6% from user interface; and 3% from code inspection. Classification on trigger attribute shows that over 50% of the escapes would have been found in test by increasing function test for coverage and variation. While coverage indicates the customer executed a single function with no or at most one set of parameters to uncover defects, variation is used when the customer uncovered a defect through varying input parameters or using boundary conditions with a single function. For example, center aligning feature worked find except when double wide spacing was chosen. In this case, the customer varied the parameter which indicated the amount of spacing in order to expose this defect. Finally the impact attribute was analyzed to determine the impact of the failure to the customer. The results show that over 60% of the failures had an impact of capability indicating the code simply did not implement the functionality the way it was specified as opposed to an impact of performance, maintenance, or reliability.


  
START Conference Manager (V2.52.6)
Maintainer: mark.sherriff@ncsu.edu