Memoized Forward Computation of Dynamic Slices

Abstract

Forward computation of dynamic slices is necessary to support interactive debugging and online analysis of long running programs. However, the overhead of existing forward computing algorithms limits their use to nonprocessing intensive applications. Recent empirical studies have shown that slices tend to reoccur often during execution. This paper presents a new forward computing algorithm for dynamic slicing, which is based on the stronger assumption that the same set union operations need to be performed repeatedly during slice computation. We present the results of an empirical study contrasting the performance of our new algorithm to the performance of a basic forward computing algorithm that unconditionally merges slices influencing an executing statement. The results indicate that the new algorithm is substantially faster than the basic algorithm and often requires significantly less memory.

Exploiting the empirical characteristics of program dependences for improved forward computation of dynamic slices

“…This paper extends the work in Masri et al (2006) by: (1) improving the space efficiency of the original algorithm; (2) optimizing the set union operations involving triples and quadruples of slices in addition to those involving pairs of slices; (3) providing an empirical study that shows several useful characteristics of dynamic slices; (4) implementing variants of the roBDD-based algorithm presented in and comparing their performance to the performance of the presented memoization-based algorithm; and (5) using additional subject programs and larger data sets. Next we survey previous work related to dynamic slicing.…”

“…The memoization-based algorithm we presented in Masri et al (2006) made the following assumption: When computing a slice at a given statement, there is a high probability that the set union operations that need to be performed on pairs of slices have already been performed during previous slice computations. In this section we empirically evaluate this assumption and identify other characteristics of program dependences that we exploit in Section 4 to improve the forward computation of dynamic slices, but first we describe the subject programs and test suites used in this study.…”

“…We modified the Dynflow tool and algorithms presented in Masri et al (2006) to conduct our experiments using the XML parser Xerces 1.3 (52,528 lines of code), the XML pretty printer JTidy 3 (9,153 lines of code), the Servlet engine jigsaw 2.0.5 (64,336 lines of code), the Servlet/JSP engine Tomcat 3.0 (21,762 lines of code), the Servlet/JSP engine Tomcat 3.2.1 (26,516 lines of code), and the XML parser NanoXML 2 for Java (7,646 lines of code). Each subject program was executed on numerous inputs of varied types and complexity.…”

Algorithms and tool support for dynamic information flow analysis

“…Note that in [30] we describe an optimization technique that drastically improves the performance of the dynamic slicing algorithm presented in this paper; we will not discuss or extend it here due to space limitations.…”

Application-based anomaly intrusion detection with dynamic information flow analysis

“…We used an updated version of this tool, which is called DynFlow, in the case studies described in Section 6, to implement fine-grained DIFA in support of information flow anomaly detection. The basic functions of DynFlow are: detecting violations of information flow policies; recording information flow profiles of executions; and computing dynamic slices to aid debugging (Masri et al, 2006). Only the second capability was employed in the case studies.…”