Instrumenting CMSSW I/O

This page discusses how to instrument CMSSW I/O at the Linux kernel in order to better understand how the application interacts with the block layer. The goals are:
  1. Provide insights into how CMSSW (and hence ROOT) I/O's usage of various system calls results in disk activity.
  2. Identify potential performance bottlenecks.
  3. Experiment with potential solutions.

Pages in this guide

BlockTrace, a set of scripts to record and visualize Linux kernel I/O.

  1. Using BlockTrace and SystemTap on your system. How to run BlockTrace on your systems in order to repeat the measurements I have made yourself.
  2. BlockTrace graphs. How to understand and read the BlockTrace graphs I will be showing.
  3. User-level SystemTap tracing. Correlating between user-level events and kernel-level events.

Findings for CMSSW

  1. CMSSW with no cache. This is the out-of-the box behavior of CMSSW.
  2. CMSSW with readHint=application-only. This turns on TTreeCache, but does I/O in synchronous mode.
  3. CMSSW with readHint=storage-only. This turns on TTreeCache with I/O in asynchronous mode.
  4. CMSSW with the 2-cache scheme. This turns on the 2-cache scheme, as described by CmsIOWork2.
Topic attachments
I Attachment History Action Size Date Who Comment
PNGpng trace.png r1 manage 49.1 K 2010-03-16 - 16:46 BrianBockelman  
Edit | Attach | Watch | Print version | History: r5 < r4 < r3 < r2 < r1 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r5 - 2010-05-13 - BrianBockelman
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    Main All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2022 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
or Ideas, requests, problems regarding TWiki? use Discourse or Send feedback