Performance tools are critical for realizing the computational potential of current and future high-end computing systems. For existing systems like the Cray XT3 at Oak Ridge National Laboratory, performance tools can provide an indication of a program's current behavior on a given system, an estimation of the achievable performance on the system, and insight into why the observed performance does not match the achievable performance. For systems that do not yet exist, tools can provide guidance to system designers and application developers allowing them to reach desired performance levels. Performance and performance modeling tools are especially important for achieving the potential of multi-paradigm computing systems, an emerging class of systems that use non-traditional processors such as FPGAs and GPUs in addition to traditional processors.
We present recent tool activity within the Future Technologies Group at Oak Ridge National Laboratory, including research into scalable performance tool techniques, communication and hardware counter trace libraries for MPI applications, trace analysis infrastructure, and multi-paradigm computing support.