The Epilog tracing library (part of the KOJAK toolset [1,2]) has been ported to the Microsoft Compute Cluster platform. The library can capture traces from MPI applications written in either FORTRAN or C/C++. The traces can be used to visualize and analyze the message passing behavior in Vampir (Intel Trace Analyzer) after converting traces to VTF format and to automatically search for patterns of inefficient execution with KOJAK/Scalasca [3].

Compared to version 2.1, KOJAK 2.2 supports MPI-2 1-sided communication, SHMEM 1-sided communication (AIX TurboSHMEM only), and holistic hardware counter analysis. Moreover, it includes a configure script for improved installation, can generate traces in the Open Trace Format (OTF), and allows conversion into the Paraver trace format. In addition, CUBE should now also build with more recent versions of wxGTK and GTK2.

This third beta release includes a configure script for improved installation, support for generating traces in the Open Trace Format (OTF), and a conversion tool to translate traces into the Paraver trace format. Also, CUBE should now build with more recent versions of wxGTK and GTK2.

This second beta release includes support for measurement and analysis of MPI-2 1-sided communication, support for measurement and analysis of SHMEM 1-sided communication (AIX TurboSHMEM only), support for holistic hardware counter analysis, an experimental perturbation-compensation filter, improved support for IBM BlueGene/L platform, and a new New-BSD-style license.

This new beta release includes important new features: support for measurement of MPI-2 1-sided communication, limited support for analysis of MPI-2 1-sided communication, and improved support for the IBM BlueGene/L platform.

This version includes important bug fixes for the 2.1 release.

This release comes with CUBE 2.0, a new utility for scalable off-line time synchronization, and a new r/w interface for EPILOG traces.

This release includes the new performance-algebra utilities for cross-experiment performance analysis and a new topology display.

The new CUBE beta release includes a very powerful tool for multi-experiment analysis that can be used to compare the effects of different optimization strategies or to integrate data from different monitoring tools.

KOJAK 2.1b comes with CUBE 2.0b and its powerful tools for multi-experiment analysis. Also included are capabilities to record virtual topologies as part of the trace file.

KOJAK 2.0 is the new stable release of KOJAK. It includes new bug fixes plus preliminary support for IBM BlueGene/L.

KOJAK 2.0 beta 2 contains several bug fixes, preliminary support of Cray X1, and the package already includes the CUBE display for simplified installation. For more information please visit the download page.

KOJAK 2.0 beta includes a faster analysis component and produces output for the CUBE display. It is based on an optimized search strategy taking advantage of specialization relationships between execution patterns and, thus, provides a significant runtime improvement over the old version. Additional speedup comes from the fact that it is now implemented in C++ and no longer in Python. For more information please visit the download page.

CONE is a call-graph profiler for MPI applications on AIX Power4 platforms, which maps hardware-counter data collected via the PAPI library onto the full call graph including line numbers. CONE is based on a call-graph tracking technique developed at IBM's Advanced Computing Technology Center (ACTC) which determines the current call path at run time while introducing only constant overhead per call.

CUBE is a generic presentation component suitable for displaying a wide variety of performance metrics for parallel programs including message-passing and shared-memory applications. CUBE allows interactive exploration of a multidimensional performance space in a scalable fashion. Scalability is achieved in two ways: hierarchical decomposition of individual dimensions and aggregation across different dimensions. All performance metrics are uniformly accommodated in the same display and thus provide the ability to easily compare the effects of different kinds of performance behavior. For more information and download please refer to the CUBE web page.