Interactive tool for visualizing performance data in real-time to enable adaptive performance optimization and feedback

摘要

An interactive tool is disclosed for visualizing performance data in real-time to enable adaptive performance optimization for an application running on a massively parallel computer system. The interactive tool may be used to visualize network congestion (and other) performance counters for an application as it runs on the parallel system in real-time. Further, a developer may use the interactive tool to experiment with various tuning options and optimization approaches on-the-fly.

主权项

1. A method of generating a visual representation of performance data for an application running on a plurality of compute nodes of a parallel computing system, wherein instances of the application communicate with one another using a first network, comprising:
receiving a value for each of one or more dedicated hardware performance counters present on the plurality of compute nodes, wherein the values are received using a second network that is separate from the first network and without disrupting the performance of the application, and wherein a first one of the one or more dedicated hardware performance counters is configured to count a number of packets that pass through a plurality of network ports coupled to one of the plurality of compute nodes; determining both a maximum value and a minimum value for each dedicated hardware performance counter, wherein the determined maximum value represents a maximum value for the dedicated hardware performance counter across all of the plurality of compute nodes, and wherein the determined minimum value represents a minimum value for the dedicated hardware performance counters across all of the plurality of compute nodes; for each of the plurality of compute nodes, determining a weighted color value relative to the received values for the other compute nodes in the plurality of compute nodes, wherein the weighted color value is based on the received value for the respective compute node, the respective maximum value for the dedicated hardware performance counter, and the respective minimum value for the dedicated hardware performance counter; and generating a three-dimensional visual display representing the plurality of compute nodes, wherein the three-dimensional visual display depicts a network topology of the plurality of compute nodes and is colored based on the weighted color values determined for the plurality of compute nodes.