DATA CENTER THERMAL MANAGEMENT

• US 20130166241A1
• Filed: 12/23/2011
• Published: 06/27/2013
• Est. Priority Date: 12/23/2011
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: HAMANN, HENDRIK F., LI, HONGFEI
• Granted Patent: US 9,258,932 B2
• Time in Patent Office: Days
• US Class Current: 702/99
• CPC Class Codes: G01K 7/42 Circuits effecting compensation of thermal inertia; Circuits for predicting the stationary value of a temperature
• CPC Class Codes: G06F 1/206 comprising thermal management
• CPC Class Codes: G06Q 10/06 Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
• CPC Class Codes: H05K 7/20836 Thermal management, e.g. server temperature control
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

Historical high-spatial-resolution temperature data and dynamic temperature sensor measurement data may be used to predict temperature. A first formulation may be derived based on the historical high-spatial-resolution temperature data for determining a temperature at any point in 3-dimensional space. The dynamic temperature sensor measurement data may be calibrated based on the historical high-spatial-resolution temperature data at a corresponding historical time. Sensor temperature data at a plurality of sensor locations may be predicted for a future time based on the calibrated dynamic temperature sensor measurement data. A three-dimensional temperature spatial distribution associated with the future time may be generated based on the forecasted sensor temperature data and the first formulation. The three-dimensional temperature spatial distribution associated with the future time may be projected to a two-dimensional temperature distribution, and temperature in the future time for a selected space location may be forecasted dynamically based on said two-dimensional temperature distribution.

Claims

1. A method of data center thermal management, comprising:

receiving historical high-spatial-resolution temperature data;

deriving a first formulation based on the historical high-spatial-resolution temperature data for determining a temperature at any point in 3-dimensional space as a function of physical observables;

receiving dynamic temperature sensor measurement data in real-time measured by a plurality of sensors at respective plurality of sensor locations;

calibrating the dynamic temperature sensor measurement data based on a calibration model using the historical high-spatial-resolution temperature data and sensor temperatures at a corresponding historical time and location;

forecasting sensor temperature data at said plurality of sensor locations for a future time based on the calibrated dynamic temperature sensor measurement data;

generating a three-dimensional temperature spatial distribution associated with the future time based on the forecasted sensor temperature data and the first formulation;

projecting the three-dimensional temperature spatial distribution associated with the future time to a two-dimensional temperature distribution; and

forecasting temperature in the future time for a selected space location dynamically based on said two-dimensional temperature distribution.

2. The method of claim 1, wherein the forecasting sensor temperature data includes generating a time series forecasting model to predict said sensor temperature data at said plurality of sensor locations for the future time.

3. The method of claim 2, wherein the time series forecasting model includes:

y*(s_i,h_l,t+1)−

φ

y*(s_i,h_l,t+1−

b)=ρ

(y*(s_i,h_l,t)−

φ

y*(s_i,h_l,t−

b))+ε

_i+1+θ

ε

_t,

wherein y*(s_ih_l,t)=y(s_i,h_l,t)−

Xβ

is a residual after removing external factor impact, b refers to seasonal cycle, φ

captures temporal correlation for seasonal term which indicates a dependence of temperature at time t on temperature at time t−

b, ρ

captures first-order temporal correlation which indicates a dependence of the temperature at time t on temperature at previous time t−

1, and θ

considers a correlation of moving average term which indicates a temporal dependence among random error terms.

4. The method of claim 1, wherein the forecasting temperature in the future time for a selected space location dynamically includes performing 2-d kriging spatial interpolation of said two-dimensional temperature distribution.

5. The method of claim 4, wherein the performing 2-d kriging spatial interpolation includes:

6. The method of claim 1, wherein the first formulation is a curve fitting model.

7. The method of claim 6, wherein the first formulation includes:

8. The method of claim 1, wherein the historical high-spatial-resolution temperature data includes MMT data taken in sparse periods of time.

9. The method of claim 1, wherein the dynamic temperature sensor measurement data in real-time is measured by a plurality of sensors sparsely located at respective plurality of sensor locations.

10. A computer readable storage medium storing a program of instructions executable by a machine to perform a method of data center thermal management, comprising:

receiving historical high-spatial-resolution temperature data;

deriving a first formulation based on the historical high-spatial-resolution temperature data for determining a temperature at any point in 3-dimensional space as a function of physical observables;

receiving dynamic temperature sensor measurement data in real-time measured by a plurality of sensors at respective plurality of sensor locations;

calibrating the dynamic temperature sensor measurement data based on a calibration model using the historical high-spatial-resolution temperature data and sensor temperatures at a corresponding historical time and location;

forecasting sensor temperature data at said plurality of sensor locations for a future time based on the calibrated dynamic temperature sensor measurement data;

generating a three-dimensional temperature spatial distribution associated with the future time based on the forecasted sensor temperature data and the first formulation;

projecting the three-dimensional temperature spatial distribution associated with the future time to a two-dimensional temperature distribution; and

forecasting temperature in the future time for a selected space location dynamically based on said two-dimensional temperature distribution.

11. The computer readable storage medium of claim 10, wherein the forecasting sensor temperature data includes generating a time series forecasting model to predict said sensor temperature data at said plurality of sensor locations for the future time.

12. The computer readable storage medium of claim 11, wherein the time series forecasting model includes:

y*(s_i,h_l,t+1)−

φ

y*(s_i,h_l,t+1−

b)=ρ

(y*(s_i,h_l,t)−

φ

y*(s_i,h_l,t−

b))+ε

_i+1+θ

ε

_t,

wherein y*(s_ih_l,t)=y(s_i,h_l,t)−

Xβ

is a residual after removing external factor impact, b refers to seasonal cycle, φ

captures temporal correlation for seasonal term which indicates a dependence of temperature at time t on temperature at time t−

b, ρ

captures first-order temporal correlation which indicates a dependence of the temperature at time t on temperature at previous time t−

l, and θ

considers a correlation of moving average term which indicates a temporal dependence among random error terms.

13. The computer readable storage medium of claim 10, wherein the forecasting temperature in the future time for a selected space location dynamically includes performing 2-d kriging spatial interpolation of said two-dimensional temperature distribution.

14. The computer readable storage medium of claim 13, wherein the performing 2-d kriging spatial interpolation includes:

15. The computer readable storage medium of claim 10, wherein the first formulation is a curve fitting model.

16. The computer readable storage medium of claim 15, wherein the first formulation includes:

17. The computer readable storage medium of claim 10, wherein the historical high-spatial-resolution temperature data includes MMT data taken in sparse periods of time.

18. The computer readable storage medium of claim 10, wherein the dynamic temperature sensor measurement data in real-time is measured by a plurality of sensors sparsely located at respective plurality of sensor locations.

19. A system for data center thermal management, comprising:

a knowledge-base including a first formulation built based on historical high-spatial-resolution temperature data, the first formulation for determining a temperature at any point in 3-dimensional space as a function of physical observables;

a calibration module for calibrating dynamic temperature sensor measurement data measured by a plurality of sensors at respective plurality of sensor locations, the calibrating performed based on a calibration model using the historical high-spatial-resolution temperature data and sensor temperatures at a corresponding historical time and location;

a first forecast module for forecasting sensor temperature data at said plurality of sensor locations for a future time based on the calibrated dynamic temperature sensor measurement data;

a second forecast module for generating a three-dimensional temperature spatial distribution associated with the future time based on the forecasted sensor temperature data and the first formulation, the second forecast module further projecting the three-dimensional temperature spatial distribution associated with the future time to a two-dimensional temperature distribution, the second forecast module further for forecasting temperature in the future time for a selected space location dynamically based on said two-dimensional temperature distribution.

20. The system of claim 19, wherein the first forecast module forecasts the sensor temperature data based on a time series forecasting model to predict said sensor temperature data at said plurality of sensor locations for the future time.

21. The system of claim 19, wherein the second forecast module forecasts the temperature in the future time for the selected space location dynamically based on performing a 2-d kriging spatial interpolation of said two-dimensional temperature distribution.

22. The system of claim 19, wherein the first formulation is a curve fitting model.