Correcting the Dynamic Call Graph Using Control-Flow Constraints

Abstract

Abstract. To reason about programs, dynamic optimizers and analysis tools use sampling to collect a dynamic call graph (DCG). However, sampling has not achieved high accuracy with low runtime overhead. As object-oriented programmers compose increasingly complex programs, inaccurate call graphs will inhibit analysis and optimizations. This paper demonstrates how to use static and dynamic control flow graph (CFG) constraints to improve the accuracy of the DCG. We introduce the frequency dominator (FDOM), a novel CFG relation that extends the dominator relation to expose static relative execution frequencies of basic blocks. We combine conservation of flow and dynamic CFG basic block profiles to further improve the accuracy of the DCG. Together these approaches add minimal overhead (1%) and achieve 85% accuracy compared to a perfect call graph for SPEC JVM98 and DaCapo benchmarks. Compared to sampling alone, accuracy improves by 12 to 36%. These results demonstrate that static and dynamic control-flow information offer accurate information for efficiently improving the DCG.

Adaptive Correction of Sampling Bias in Dynamic Call Graphs

“…The correction technique of using call density (Call density) shows the improvement of 0.5% on average, and the correction using both call density and sampling latency (Call density and sampling latency) improves the performance up to 1% on average. While this modest improvement is consistent with the prior work [Arnold and Grove 2005;Lee et al 2007], some relatively large and complex benchmarks show potential of using high-accuracy dynamic call graphs. For instance, our corrected dynamic call graphs improve the performance of bloat, jython, jack, lusearch, and xalan by 6.1%, 1.8%, 1.5%, 1.5%, and 1.5%.…”

“…While our revised inline oracle resorts to static heuristics to handle a few pathological cases, we strongly believe that profile-guided inlining oracles should weaken static heuristics as much as possible as low overhead sampling and correction techniques will evolve and mature [Arnold and Grove 2005;Lee et al 2007;Chen et al 2010;Wu et al 2013;Nakaike et al 2014]. The performance improvement from static inlining of inactive calls to trivial methods would be either marginal or unmeasurable in a registerrich environment (e.g., ×86-64).…”

“…Our experimental results demonstrate the effectiveness of our antibias techniques on software-based sampling techniques, and we believe that our techniques would also work well for the hardware-based approaches as long as their sampling actions are triggered by wall clock timers or CPU cycle counters. Lee et al [2007] discover static and dynamic control-flow constraints and reconcile the sampling errors in the dynamic call graph with these constraints. Chen et al [2010] apply supervised learning techniques to hardware-based sampling to improve the accuracy of basic-block-level sampled profiles.…”

Simplifying the Structural Complexity of Software Systems

“…Nevertheless, most of them are based on using "fan-in analysis techniques." Therefore, they are depending on either dependency graphs or call graphs, which suffer from precision issues [18]. Notwithstanding, there is not much research depending on the original source of data utilizing a specific technique, for instance, dynamic coupling measurements [14].…”

A survey on Test Suite Reduction frameworks and tools

“…Jtop automatically extracts the static call graph in order to generate nonredundant test cases. Future tools can further improve the capability of Jtop by constructing a dynamic call graph (Lee et al, 2007), which extracts an exact record of program execution frequencies of call relationship based on the profile information. Ultimately, it is beneficial in optimizing overall program under test behavior.…”