Browse Prior Art Database

Method and System for Optimizing Energy Usage in Server Clusters

IP.com Disclosure Number: IPCOM000202421D
Publication Date: 2010-Dec-15
Document File: 3 page(s) / 81K

Publishing Venue

The IP.com Prior Art Database

Abstract

A method and system for optimizing energy usage in server clusters is disclosed.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 47% of the total text.

Page 01 of 3

Method and System for Optimizing Energy Usage in Server Clusters

Disclosed is a method and system for optimizing energy usage in server clusters. Historical data corresponding to energy utilization of a server cluster is monitored and analyzed. Based on the historical data future energy utilization of the server cluster is predicted and one or more energy optimization policies are provided. The one or more energy optimization policies are evaluated with respect to energy savings. Thereafter, servers are categorized into one or more sets of server based on one or more attributes and the one or more optimization policies are implemented on the one or more sets of servers.

The method and system disclosed herein, monitors workload of servers in a cluster and gathers the workload information at periodic intervals of time (for example, every hour). The workload of a server is modeled based on capacity of three physical attributes of the server, namely, CPU, memory, and disk. Accordingly, the server workload may be represented using as a 3-tuple,Wi = (ci,di, mi

                            ), wherein Wi is the workload on the ith server, ciis a range of values denoting work performed by a CPU in the ith server, di is a range of values denoting the disk utilized for performing the work on the ith server and mi is a range of values denoting the memory allotted for performing the work in the ith

server. Accordingly, an aggregated workload for a server cluster with n servers may be

determined by summing Wi over all servers of the server cluster, i.e. . Further, a utilization pattern Uiof the ithserver is given by Ui = (Wi, start, end

)

(This page contains 01 pictures or other non-text object)

                                            is also determined for each server. In this case, start is the start date and time of the utilization pattern of the ithserver and end is the end date and time of the utilization pattern of the ithserver. Utilization pattern represents the workload on a server during a specified duration of time. Thereafter, the method and system models the power consumption of a server

cluster as . In this case, Pi is the power consumed by the ithserver, Bi is the base power consumed by ith server when the server is in an idle mode, Mr,I is a measure of the power of a server resource r of ith server when the server is being fully utilized, Rr,i is the utilization of the server resource r on ith server, and Cr,i is the capacity of the server resource r on ith server.

Based on the modeled power consumption, the method and system disclosed herein creates a model for a predicted workload for the server cluster based on historic workload data. The predicted workload is modeled as a "Potluck Problem". An Autocorrelation Function (ACF) and a Partial Autocorrelation Function (PACF) may be used for determining statistical properties and trends of the historic workload data, which is available as a time-series data. The trends are used for building one or more predictors for workload in the server cl...