Detecting inefficiently-used containers to avoid bloat

  • Authors:

  • Guoqing Xu;Atanas Rountev

  • Affiliations:

  • Ohio State University, Columbus, OH, USA;Ohio State University, Columbus, OH, USA

  • Venue:
  • PLDI '10 Proceedings of the 2010 ACM SIGPLAN conference on Programming language design and implementation
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

Runtime bloat degrades significantly the performance and scalability of software systems. An important source of bloat is the inefficient use of containers. It is expensive to create inefficiently-used containers and to invoke their associated methods, as this may ultimately execute large volumes of code, with call stacks dozens deep, and allocate many temporary objects. This paper presents practical static and dynamic tools that can find inappropriate use of containers in Java programs. At the core of these tools is a base static analysis that identifies, for each container, the objects that are added to this container and the key statements (i.e., heap loads and stores) that achieve the semantics of common container operations such as ADD and GET. The static tool finds problematic uses of containers by considering the nesting relationships among the loops where these semantics-achieving statements are located, while the dynamic tool can instrument these statements and find inefficiencies by profiling their execution frequencies. The high precision of the base analysis is achieved by taking advantage of a context-free language (CFL)-reachability formulation of points-to analysis and by accounting for container-specific properties. It is demand-driven and client-driven, facilitating refinement specific to each queried container object and increasing scalability. The tools built with the help of this analysis can be used both to avoid the creation of container-related performance problems early during development, and to help with diagnosis when problems are observed during tuning. Our experimental results show that the static tool has a low false positive rate and produces more relevant information than its dynamic counterpart. Further case studies suggest that significant optimization opportunities can be found by focusing on statically-identified containers for which high allocation frequency is observed at run time.