Classroom instruction combines an overview of the basics of debugging an application with step-by-step lab exercises and tutorials. This class offers an efficient way to streamline and simplify the task of debugging parallel codes, no matter how complex, using the TotalView Source Code Debugger.
Workshop Outline
* Startup and Source Code Navigation
* Data and Process Control
* Using Action Points
* Data Monitoring and Visualization
* Asynchronous Control for Parallel Applications
* Viewing Data Across Processes and Threads
* Strategies for Memory Debugging
* Leak Detection and Memory Corruption Detection
* Memory Events/Error Notification
* Memory Reporting
* Scripting and batch (non-interactive) debugging
* Remote display debugging
* Reverse Debugging with ReplayEngine (for Linux x86, x86-64 users)
Format
Instructor-led (combined lecture and hands-on lab)
The Hands-on lab utilizes a pre-packaged, bootable liveCD that runs on any Intel-based system. The use of the CD will not change anything on the users hard disk, and no network connectivity is required. The lab is also designed to be self-paced, when used with the lab manual, so that exercises begun in the lab session can be continued and/or repeated by the users afterward at their leisure.
Prerequisites
Attendees should be familiar with the basics of parallel programming with threads (pthreads, OpenMP, etc), and/or multi-process programming, (MPI, OpenMPI, UPC, etc.) with C/C++ or FORTRAN. A basic understanding of cluster computing would also be beneficial.
The Training CD contains:
• Bootable version of Ubuntu Linux,
• TotalView Debugger evaluation software
• Product documentation PDFs
• TotalView Lab Manual PDF
In addition to the training CD, class participants receive hard copies of the course manual, the answer key to exercise questions and a set of Totalview Quick Reference guides.