It is my pleasure to announce that we are starting the beta for TotalView 8.15.

The 8.15.0 release focus is on scalability and performance improvements at scale. This release of TotalView uses a new scalable infrastructure when you are debugging on Linux, Blue Gene/Q or Cray platforms. This new infrastructure supports efficient broadcast and reduction operations and is a quantum leap in TotalView scalability. TotalView can now be used across hundreds of thousands of processes and millions of threads. In this release, we’ve tuned TotalView to take advantage of that broadcast and reduction to accelerate the performance of many complex operations.

The most notable change you’ll see is with startup performance. When debugging jobs of 1,000 or more MPI processes, we’ve shortened the amount of time between when you ask TotalView to launch the job and when the job has started, ready for you to begin debugging. At larger scales, you’ll see a larger impact.

The large scale improvements in version 8.15 build upon recent enhancements for debugging applications with large and complex symbol tables. We’ve made changes to the way TotalView handles symbol table searches and have measured significant improvements in responsiveness and performance when setting breakpoints, looking up and stepping into functions, hovering over variable names, and other common activities with applications that make heavy use of C++ templates and dynamically-loaded shared objects.

And, in this new release, Memory Debugging startup time has been optimized, with memory debugging now usable for thousands and tens of thousands of processes.

A few notes for the 8.15 release:
  • Many memory debugging operations after startup have not been optimized specifically, so you may find it necessary to focus on a single process to get acceptable performance when memory debugging at large scale.
  • Additionally, the processes tab on the process window is gone. Its dynamic always updated display of process state was simply not something that we were able to scale.
  • The root window now displays an aggregated tree of information about the state of the process and threads that you are debugging, replacing the previous list-based view of processes.
  • You can now choose what properties to use when aggregating. The root window can still be used to focus on a particular process.
  • Improved delayed symbol processing. Significantly improved performance when processing debug symbols on large executables, which tends to cause delays as all the information needed is acquired when performing various operations in a debugging session.
  • There are two compiler updates in this release: Intel 15.0 and PGI C++ 14.4.

If you are interested in participating in the TotalView 8.15 beta release, please email

with the following information:

--How long have you used TotalView?
--Will you be able to test TotalView at large scale? If so, how large a job can you test and can you provide timing information to us?
--Will you be able to test TotalView on programs which either contain large numbers of shared objects or are large statically linked executables?
--Will you be able to test Memory Debugging?
--On what platforms do you plan to test TotalView?
--Do you have any feedback on beta and prerelease candidate testing?

If you do participate we will appreciate your interest and input.

Happy Holidays!

-Chris Gottbrath