Methods of achieving cognizant power management

Abstract

A system and method of increasing the efficiency of overall power utilization in data centers by integrating a power management approach based on a comprehensive, dynamic model of the data center created with integrated environmental and computational power monitoring to correlate power usage with different configurations of business services utilization, with the techniques of CPU level power management.

What is claimed is: 
     
         1 . A method for cognizant power management in a distributed computing system having a plurality of processing units, comprising the steps of:
 generating a predictive behavior power utilization model based on first values of process level information associated with the processing units and first values of environmental metrics for the computing system;   predicting a power utilization of the computing system according to the predictive behavior power utilization model and a selected computational load to be distributed among the processing units;   distributing the computational load to the processing units computing system based on based on the predictive behavior power utilization model;   determining an actual power utilization by gathering second values for the process level information and the environmental metrics; and   updating the predictive behavior power utilization model according to actual power utilization when a difference between the actual power utilization and the predicted power utilization exceeds a predetermined threshold.   
     
     
         2 . The method of  claim 1 , wherein the plurality of processing units includes at least one of a plurality of central processing units, a plurality of functional units within central processing units, and a plurality of hypervisors managing virtual machines. 
     
     
         3 . The method of  claim 2 , further comprising the step of:
 setting a state in at least one of the plurality of processing units based on the predictive behavior power utilization model before carrying out the distributing step, wherein the state is one of a plurality of performance states.   
     
     
         4 . The method of  claim 3 , wherein setting one of a plurality of performance states includes changing the frequency of a processing unit. 
     
     
         5 . The method of  claim 3 , wherein setting one of a plurality of performance states includes changing the voltage of a processing unit. 
     
     
         6 . The method of  claim 3 , wherein setting one of a plurality of performance states includes changing the frequency and the voltage of a processing unit. 
     
     
         7 . The method of  claim 3 , wherein one of a plurality of performance states includes at least a first performance state and a second performance state, wherein the first performance state can bear a greater computational load than the second state. 
     
     
         8 . The method of  claim 3 , wherein the state is one of a plurality of sleep states. 
     
     
         9 . The method of  claim 8 , wherein one of a plurality of sleep states includes at least a first sleep state and second sleep state, wherein the first sleep state has a lower wake up time than the second sleep state. 
     
     
         10 . The method of  claim 1 , further comprising the step of:
 gathering the first values of the process level information from the plurality of processing and the first values of environmental metrics for the computing system before carrying out the predictive step.