Using complete system simulation to characterize SPECjvm98 benchmarks

  • Authors:

  • Tao Li;Lizy Kurian John;Vijaykrishnan Narayanan;Anand Sivasubramaniam;Jyotsna Sabarinathan;Anupama Murthy

  • Affiliations:

  • -;-;-;-;-;-

  • Venue:
  • Proceedings of the 14th international conference on Supercomputing
  • Year:
  • 2000

Quantified Score

Hi-index 0.02

Visualization

Abstract

Complete system simulation to understand the influence of architecture and operating systems on application execution has been identified to be crucial for systems design. While there have been previous attempts at understanding the architectural impact of Java programs, there has been no prior work investigating the operating system (kernel) activity during their executions. This problem is particularly interesting in the context of Java since it is not only the application that can invoke kernel services, but so does the underlying Java Virtual Machine (JVM) implementation which runs these programs. Further, the JVM style (JIT compiler or interpreter) and the manner in which the different JVM components (such as the garbage collector and class loader) are exercised, can have a significant impact on the kernel activities.To investigate these issues, this research uses complete system simulation of the SPECjvm98 benchmarks on the SimOS simulation platform. The execution of these benchmarks on both JIT compilers and interpreters is profiled in detail, to identify and quantify where time is spent in each component. The kernel activity of SPECjvm98 applications constitutes up to 17% of the execution time in the large dataset and up to 31% in the small dataset. The average kernel activity in the large dataset is approximately 10%, in comparison to around 2% in four SPECInt benchmarks studied. Of the kernel services, TLB miss handling is the most dominant in all applications. The TLB miss rates in the JIT compiler, dynamic class loader and garbage collector portions of the JVM are individually analyzed. In addition to such execution profiles, the ILP in the user and kernel mode are also quantified. The Java code is seen to limit exploitable parallelism and aggressive instruction issue is seen to be less efficient for SPECjvm98 benchmarks in comparison to SPEC95 programs. Also, the kernel mode of execution does not exhibit as much ILP as the user mode.