ParaTrac is a profilier for parallel/distributed applications. It enables user to investigate their applications by non-intrusive tracing and auto-generated informative profiles.

Features

• User-level tracing for both file system call and process statistics
• Fine-grained profiles for application investigation

Profiling Data-Intensive Workflows

Data-intensive workflow are current target application of ParaTrac.

• Low-level I/O profiles: General application behavior, I/O pattern/prediction, bottlenecks of underlying I/O systems.
• Workflow profiles: Data-process interactions, processing pattern, inefficient scheduling detection, workflow debugging.

Here is some exemplary profiles of data-intensive workflows.

Future Directions

• Memory reference profiling
• Profile replaying

Publications

Nan Dun, Kenjiro Taura and Akinori Yonezawa. ParaTrac: A Fine-Grained Profiler for Data-Intensive Workflows. In Proceedings of The ACM International Symposium on High Performance Distributed Computing (HPDC 2010), to appear, Chicago, Illinois, June 2010. (pdf, Slides)

Nan Dun, Kenjiro Taura and Akinori Yonezawa. Fine-Grained Profiling for Data-Intensive Workflows. In Proceedings of the 10th IEEE International Symposium on Cluster Computing and the Grid (CCGrid 2010). Best Poster, Melbourne, Australia, May 2010. (pdf, A1 Poster)