Profiling a Many-core Neuromorphic Platform

This paper presents a unified profiling platform for a many-core machine that gives realtime information about the current state of the system. Such information is valuable for the observation and evaluation of running programs on the machine, as well as the health status of the machine itself. The information from the profiler can also be used for tuning the system's kernel operational parameters to maintain its performance and reliability. In our work, the profiler framework is developed for SpiNNaker, a many-core neuromorphic platfrom. The profiler framework provides realtime information such as power consumption, chip temperature, processor frequency, core utilization, and network connectivity. Many researchers and institutions have been using SpiNNaker not only for simulating spiking neural networks, but also for general purpose and energy-aware computing. Therefore, it is important to provide a reliable and comprehensive profiling platform for helping SpiNNaker users in developing their programs.