Monitoring software using property-aware program sampling

Abstract

<p>Monitoring or profiling programs provides us with an</p> <p>understanding for its further improvement and analysis.</p> <p>Typically, for monitoring or profiling, the program is instrumented</p> <p>to execute additional code that collects necessary data.</p> <p>A problem is that program instrumentation is often reported to</p> <p>cause between 10% and 390% time and space overhead.</p> <p>A number of techniques based on statistical sampling</p> <p>have been proposed to reduce the instrumentation overhead.</p> <p>Statistical sampling based instrumentation techniques,</p> <p>although effective in reducing the overall overhead,</p> <p>often lead to poor coverage or less accurate results.</p> <p>In this work, we present a profiling technique based</p> <p>on property-aware program sampling.</p> <p>The key ideas are (i) to use program slicing to narrow</p> <p>down the scope of instrumentation to the sections</p> <p>of program relevant to the property of interest,</p> <p>(ii) to decompose large program slices into logically</p> <p>related slice fragments, and (iii) to apply statistical</p> <p>sampling on the set of slice fragments.</p>