From fca59bea770346cf1c1f9b0e00cb48a61b44a8f3 Mon Sep 17 00:00:00 2001 From: Harald Welte Date: Sun, 25 Oct 2015 21:00:20 +0100 Subject: import of old now defunct presentation slides svn repo --- .../OLS2005/kacur/kacur-abstract.tex | 29 ++++++++++++++++++++++ 1 file changed, 29 insertions(+) create mode 100644 2005/flow-accounting-ols2005/OLS2005/kacur/kacur-abstract.tex (limited to '2005/flow-accounting-ols2005/OLS2005/kacur/kacur-abstract.tex') diff --git a/2005/flow-accounting-ols2005/OLS2005/kacur/kacur-abstract.tex b/2005/flow-accounting-ols2005/OLS2005/kacur/kacur-abstract.tex new file mode 100644 index 0000000..8c744be --- /dev/null +++ b/2005/flow-accounting-ols2005/OLS2005/kacur/kacur-abstract.tex @@ -0,0 +1,29 @@ + +% Registration Profiling Java on Linux +% [2]Register/Submit Proposal John Kacur (jekacur@ca.ibm.com) + +In this paper, I will examine two profilers. +IBM's Open Source Performance Inspector and +OProfile which contains code that has been +officially accepted into the Linux Kernel. +Currently OProfile doesn't work with programs +that dynamically generate code, such as Python +and Java JITs. Various people have proposed +patches that record events in anonymously +mapped memory regions as raw virtual +addresses, instead of the usual tuple of +binary image and offset. This information can +be postprocessed by matching it with the +output generated by running a Java program +with Performance Inspector's JPROF which uses +JVMPI to record addresses of JITted methods. +In this paper, I will discuss the details of +profiling Java, specifically looking at the +inner workings of OProfile and Performance +Inspector. I will discuss problems that we +have encountered with both tools and our +attempts to resolve them. Finally, I will +demonstrate profiling a java program to show +the kind of information that can be obtained. + + -- cgit v1.2.3