Vampir meets ISC '1001
Welcome to the June edition of Vampir News, designed to keep you informed of recent developments of our performance analysis environment. This includes the tracing tool VampirTrace as well as the visualization and analysis framework of Vampir 7.
The new VampirTrace 5.9 sets focus on scalability and additional tracing features to provide an even more detailed and holistic program observation. Monitoring of NVIDIA CUDA applications now enables a first insight into accelerator usage. This functionality is based on the Vampir- Trace library tracing capability and an interpretation of CUDA API functions. It records information such as kernel execution times and data transfers between host and accelerator. The new counter plugin interface allows an easy integration of external counters without recompiling VampirTrace. In addition, to already available counter plugins such as I/O activity, CPU temperatures, and kernel events the user can specify any custom counter plugin. Furthermore, the scalability and performance of VampirTrace are improved by reducing the number of used file handles and with an automatic MPI parallel unification of local trace data.
The new Vampir 7.2 version is now available with full support for both Windows and Linux/ Unix systems. This release introduces a first version of the Performance Radar chart (Picture to the left). By now users can take advantage of this new chart which already allows to find arbitrary functions in the trace data and to perform basic counter arithmetic’s. The Performance Radar functionality will be further enhanced in the next releases to allow efficient pinpointing of performance critical sections in the application programs. Furthermore, Vampir 7.2 features the visualization of file I/O operations including a range of versatile filter options. Extended appearance settings and free selectable fonts of the performance charts greatly improve the usability of the tool. Vampir 7 Professional additionally includes the new VampirServer 2.2 version of the parallel analysis engine that enables highly scalable and efficient analysis of bulky trace data. Vampir 7 is fully compatible with previous releases of Vampir and VampirTrace. On Linux/Unix platforms, VampirTrace provides rich program monitoring options. The performance data acquisition process on Windows platforms is fully integrated into Windows HPC Server 2008.