Last Wednesday, Dr. Oscar Naim form Univ. of Wisconsin, Madison gave us a very interesting topic about the parallel Performance tools. First he explained the motivation for the Paradyn. There are:

• Profile chanllenge
• Scalability
• Automate the tuning process
• Support Heterogeneity
• Support high-level language and library

Then Dr. Naim discussed a new approach which applied in the Paradyn, called dynamic instrumentation, that is based on dynamically controlling what performance data is to be collected. Dynamic instrumentation allows data collection instructions to be inserted into an application program during runtime. Paradyn is able to look for a variety of performance problems during a single execution, while not being overwhelmed with all the available data at any one time. In addition, dynamic instrumentation that has not be inserted incurs no overhead, in contrast to instrumentation that is staticly-inserted but currently disabled.

Paradyn also provides decision support for the tool user. Paradyn assists the user in deciding which bottlenecks to look for; in fact, Paradyn can search automatically to isolate performance bottlenecks, and then explain them to the user using descriptions or visualizations.

Paradyn incorporates a new approach for describing performance information to users of high-level parallel languages; Paradyn maps performance data between multiple layers of abstraction, and user can choose to look at it in terms of high-level language constructs or low-level machine structures.