Monitoring and Reporting Proposal

From WikiOAR

Cluster Monitoring/Reporting for Greener Computing -- Validation/Configuration upon Grid5000

This project will be based upon the work done on the GSOC 2008 Energy Saving project [1]. The current energy saving module consists of:

A Prediction based energy efficient scheduling which manages the under-utilization periods of a cluster by shutting down (or suspending) computing nodes.
A new type of jobs called PowerSaving which permits users and application to perform energy saving on-demand

The first part of the work of the student for this project, will be centered on configuring and evaluating the already existing OAR code for energy saving upon Grid5000 platform. The goal of those tests are to provide a fully operational Energy Saving module upon Grid5000 platform. The configuration part demands linux system skills, kernel updates and some Perl/Ruby programming. For the evaluation of the code some bash programming skills will be needed. A small period of learning how to use and perform experiments upon Grid5000 will be necessary. Skills upon cluster and grid technologies will be useful.

The second part of this project will be centered on extending the OAR Energy Saving module. An important extension may be the development of energy monitoring and reporting commands for OAR Resource Management System. Those commands will be used to provide a detailed view of the clusters energy consumption. In more detail, it is essential:

1) to monitor system energy use by collecting specific data from machine's hardware components (Network Card, CPU, GPU, Memory, etc) to provide a specific view of power consumption upon the cluster. For this reason an extension of OarMonitor command [6], already existing in OAR, could be proposed. A Variation of PowerTop linux command [5] for clusters could be proposed.

2) to report energy use by user, project, and hardware components through visualization tools so as to understand how resources are being used and how power is being consumed for different workload scenarios. The results can be used and evaluated so as to propose actions for less energy consumption.

The experiments will be effectuated upon Grid5000 platform [4] which provides all necessary equipment for energy monitoring. This project will be in cooperation with Green-Net Project .