System and method for optimizing energy consumption by processors

Patent metadata

Field	Value
Publication number	US-9696786-B2
Application number	US-201514848632-A
Country	US
Kind code	B2
Filing date	Sep 9, 2015
Priority date	Apr 29, 2015
Publication date	Jul 4, 2017
Grant date	Jul 4, 2017

Abstract

Methods and systems for selecting an appropriate forecasting model for accurate workload prediction of a processor is disclosed. The processor is configured to monitor workload and extract workload history of the processor for a given time interval. Further, the processor is configured to create plurality of forecasting models based on the extracted workload history and apply the group of forecasting model on the extracted workload history to obtain a plurality of predicted future workload for the given time interval. Further, the processor is configured to compute an error measure of the plurality of predicted future workload in reference to an actual workload of the processor and select the appropriate forecasting model from plurality of the forecasting models having least error measure among the computed error measures for dynamically scaling frequency and voltage required by the processor and thereby optimizing energy consumption in the processor.

First claim

What is claimed is:

1. A computer implemented method for optimizing energy consumption by a processor, the method to be executed on a computer, the computer including a memory, the method comprising:
monitoring workload of said processor;
extracting workload history of said processor for a given time interval;
creating a plurality of forecasting models based on said extracted workload history;
storing said plurality of forecasting models in a repository of said memory;
applying a forecasting model from said plurality of forecasting models on said extracted workload history;
reiterating the applying step on each of said plurality of forecasting models to obtain a plurality of future workloads predicted by each of said plurality of forecasting models for the given time interval;
collating said plurality of predicted future workloads obtained from each of said plurality of forecasting models;
computing an error measure of said plurality of predicted future workloads with reference to an actual workload of said processor for said time interval;
sorting said computed error measure related to each of said plurality of forecasting models;
selecting an appropriate forecasting model among the plurality of said forecasting models based on its least error measure among said sorted error measures for optimizing energy consumption in said processor;
generating an appropriate frequency based on said predicted future workloads associated with the selected forecasting model; and
generating an appropriate voltage for said processor based on said generated appropriate frequency.

2. The method of claim 1 , wherein extracting said workload history comprises a plurality of instructions processed by said processor on hourly, daily, weekly and monthly period of time.

3. The method of claim 1 , wherein creating said plurality of forecasting models is based on varying workload conditions extracted from said workload history.

4. The method of claim 1 , wherein creating said plurality of forecasting models is based on varying workload type extracted from said workload history.

5. The method of claim 1 , wherein said plurality of forecasting models include at least one of Weighted Moving Average, Exponential Smoothing, Holt and ARIMA.

6. The method of claim 1 , wherein computing said error measure is based on MAPE or MASE.

7. A computer implemented system for optimizing energy consumption by a processor, the system comprising:
a memory coupled to said processor, wherein said memory comprises instructions which, when executed by said processor, cause said processor to:
monitor workload of said processor;
extract workload history of said processor for a given time interval;
create a plurality of forecasting models based on said extracted workload history;
store said plurality of forecasting models in a repository of said memory;
apply a forecasting model from said plurality of forecasting models on said extracted workload history;
reiterate the applying step on each of said plurality of forecasting models to obtain a plurality of future workloads predicted by each of said plurality of forecasting models for the given time interval;
collate said plurality of predicted future workloads obtained from each of said plurality of forecasting models;
compute, through an error measuring tool, an error measure of said plurality of predicted future workloads with reference to an actual workload of said processor for said time interval;
said processor configured to sort said computed error measure related to each of said plurality of forecasting models;
select an appropriate forecasting model among the plurality of said forecasting models based on its least error measure among said sorted error measures for optimizing energy consumption in said processor;
generate an appropriate frequency based on said predicted future workloads associated with the selected forecasting model; and
generate an appropriate voltage for said processor based on said generated appropriate frequency.

8. The system of claim 7 , wherein said processor is further configured to extract said workload history comprises a plurality of instructions processed by said processor on hourly, daily, weekly and monthly period of time.

9. A computer program product comprising a non-transitory computer readable storage medium having a computer readable program stored therein, wherein the computer readable program, when executed on a computing device, causes the computing device to:
monitor workload of a processor;
extract workload history of said processor for a given time interval;
create a plurality of forecasting models based on said extracted workload history;
store said plurality of forecasting models in a repository of said memory;
apply a forecasting model from said plurality of forecasting models on said extracted workload history;
reiterate the applying step on each of said plurality of forecasting models to obtain a plurality of future workloads predicted by each of said plurality of forecasting models for the given time interval;
collate said plurality of predicted future workloads obtained from each of said plurality of forecasting models;
compute, through an error measuring tool, an error measure of said plurality of predicted future workloads with reference to an actual workload of said processor for said time interval;
sort said computed error measure related to each of said plurality of forecasting models; and
select an appropriate forecasting model among the plurality of said forecasting models based on its least error measure among said sorted error measures for dynamically scaling frequency and voltage required by said processor based on the predicted future workloads associated with the selected forecasting model and thereby optimizing energy consumption in said processor.

Assignees

• Tata Consultancy Services Ltd

Inventors

• Panda Prasanta
• Singh Brajesh

Classifications

• G06F1/324 Primary

  by lowering clock frequency · CPC title

• G06F1/329

  by task scheduling · CPC title

• G06F1/3206

  Monitoring of events, devices or parameters that trigger a change in power modality · CPC title

• G06F1/3296

  by lowering the supply or operating voltage · CPC title

• G06F9/5094

  where the allocation takes into account power or heat criteria (power management in computers in general G06F1/3203; thermal management in computers in general G06F1/206) · CPC title