TAU Performance System

For the past twelve years, the TAU project has conducted research on performance tools for parallel and distributed systems. The goal of this work has mainly been the development of robust technology to meet evolving performance evaluation challenges of state-of-the-art parallel systems and applications. In particular, we have focused on problems of performance tool portability, extendability, and interoperation. The TAU performance system [9,17] is our integrated toolkit for performance instrumentation, measurement, analysis, and visualization of large-scale parallel applications. It targets a general computation model consisting of shared-memory computing nodes where contexts reside, each providing a virtual address space shared by multiple threads of execution. The model is general enough to apply to many high-performance scalable parallel systems and programming paradigms. Because TAU enables performance information to be captured at the node/context/thread levels, this information can be mapped to the particular parallel software and system execution platform under consideration. As shown in Figure 1, the TAU system supports a flexible instrumentation model that applies at different stages of program compilation and execution. The instrumentation targets multiple code points, provides for mapping of low-level execution events to higher-level performance abstractions, and works with multi-threaded, message passing, and mixed-mode parallel computation models. Different instrumentation techniques are supported, including dynamic instrumentation using the DyninstAPI [4]. All instrumentation code makes calls to the TAU measurement API to provide a common measurement model. The TAU measurement library implements performance profiling and tracing support for performance events occurring at function, method, basic block, and statement levels. Performance experiments can be composed from different measurement modules (e.g., hardware performance monitors, such as PAPI [3]) and measurements can be collected with respect to user-defined performance groups. C, C++, Fortran 77/90, OpenMP, and Java languages are supported. The TAU data analysis and presentation utilities offer text-based and graphical tools to visualize the performance data as well as bridges to third-party software, such as Vampir [11] and Paraver [12] for sophisticated trace analysis and visualization.

Figure 1: Architecture of the TAU performance system.

TAU has been ported to nearly all high-performance computing platforms and is being used extensively in the performance analysis of DOE applications. TAU is also being applied as the primary performance technology across a diverse set of code development projects, including Uintah [15], CCA [2], VTF [18], and SAMRAI [7]. Although the current set of features in the TAU performance system is quite substantial, it is important to note that users are always requesting new capabilities. The interesting research problems that arise concern how to develop new techniques to address these requests while maintaining tight integration with the rest of the TAU system. The four problems below are all representative of such endeavors.