Integration of low-level sensors with TensorBoard
Abstract
Machine learning systems require substantial computational andenergy resources, particularly for training tasks. Profilers for these systems,such as the available on the TensorBoard tool for TensorFlow, provide insightinto performance hotspots and optimization opportunities. Nevertheless, suchtools typically lack energy profiling capabilities. In this work we present anpower dissipation profiling plugin for machine learning tasks based on low-levelIPMI sensors. The plugin is integrated with TensorBoard and demonstrated withan IBM POWER machine.
References
IPMI. Intelligent platform management interface specification v2.0 rev. 1.1. [link].
ipmitool. ipmitool is a utility for managing and configuring devices that support the intelligent platform management interface. https://github.com/ipmitool/ipmitool.
Performance Application Programming Interface. Papi provides the tool designer and application engineer with a consistent interface and methodology for use of the performance counter hardware found in most major microprocessors. https://icl.utk.edu/papi/.
POWER9. Ibm processor architcture. https://www.ibm.com/br-pt/it-infrastructure/power/power9.
PowerBoard.Power profiling plugin project fortensor board. https://github.com/Unicamp-OpenPower/PowerBoard.
PowerBoardPlugin. Powerboard plugin for tensorboard. https://openpower.ic. unicamp.br/post/powerboard-plugin-for-tensorboard/.
Profiler.Tensor flow profiler to profile the execution oftensorflowcode. https://www.tensorflow.org/tensorboard/tensorboard_profiling_keras.
PyPi. The python package index. https://pypi.org/.
TensorBoard. Tensorboard: kit de ferramentas de visualização do tensorflow. https://www.tensorflow.org/tensorboard?hl=pt-br.
TensorBoard Plugin. Custom tensorboard visualization. [link].
TensorFlow. Tensorflow is an end-to-end open source platform for machine learning. https://www.tensorflow.org/.
