High-accuracy virtual sensors for computer systems

• US 7,751,910 B2
• Filed: 04/16/2007
• Issued: 07/06/2010
• Est. Priority Date: 04/16/2007
• Status: Active Grant

• Current Assignee: Oracle America, Inc. (Oracle Corporation)
• Original Assignee: Oracle America, Inc. (Oracle Corporation)
• Inventors: Cooley, John J., Gross, Kenny C., Urmanov, Aleksey M.
• Primary Examiner(s): Patel; Ramesh B
• Assistant Examiner(s): Stevens; Thomas H
• Application Number:
  US11/787,604
• Publication Number:
  US 20080255819A1
• Time in Patent Office: 1,177 Days
• Field of Search: 700/52, 702/104, 702/127, 702/90, 702/91, 702/116, 702/182
• US Class Current: 700/52
• CPC Class Codes: G06F 11/008 Reliability or availability analysis

Abstract

Embodiments of the present invention provide a system that estimates the value of a virtual sensor. The system first samples values for performance metrics using external sensors that are coupled to a system and internal sensors that are built into the system. Next, the system generates an inferential sensing model for the system from the sampled values. Then, during operation, the system samples values of performance metrics using the internal sensors and uses the inferential sensing model and the sampled values from the internal sensors to estimate the values of performance metrics for removed external sensors (i.e., virtual sensors).

Claims

1. A method for determining an estimated value for a sensor which has been removed from a system, comprising:

in a computer system;

sampling values for performance metrics using external sensors that are coupled to a system and internal sensors that are built into the system, wherein at least one internal sensor is a software sensor;

generating an inferential sensing model for the system from the sampled values; and

for one or more external sensors that have been removed from the system;

sampling values of performance metrics using the internal sensors; and

generating estimated values for performance metrics for the one or more removed external sensors, by using the sampled values from the internal sensors as inputs to the inferential sensing model.

2. The method of claim 1, wherein when at least one internal sensor is removed from the system and at least one internal sensor remains in the system, the method further comprises:

during operation, using the inferential sensing model and sampled values of performance metrics from the remaining internal sensors to generate an estimated value for a performance metric for the removed internal sensor.

3. The method of claim 1, wherein during operation, the method further comprises using the inferential sensing model to generate an estimated value for a performance metric for at least one internal sensor from sampled values of performance metrics collected from the remaining internal sensors.

4. The method of claim 1, wherein the method further comprises:

determining if the sampled values or the estimated values for the performance metrics have exceeded a threshold; and

if so, generating an alarm.

5. The method of claim 1, wherein sampling the performance metrics involves:

executing one or more test loads on the system; and

sampling the performance metrics while the test loads are executing on the system.

6. The method of claim 1, wherein generating an inferential sensing model involves using a Logical Combination of Regression Primitives (LCORP) technique, a Non-Linear Non-Parametric (NLNP) regression technique, or another statistical or regression technique to generate a mathematical model that correlates sampled values of performance metrics.

7. The method of claim 1, wherein sampling values for performance metrics further comprises using sensors that are coupled to a system selected from the group consisting of:

a chip;

a component;

a field-replaceable unit (FRU); and

a computer system.

8. The method of claim 1, wherein the performance metrics are physical system properties which can include:

a temperature, a relative humidity, a cumulative or differential vibration, a fan speed, an acoustic signal, a current, a voltage, a time-domain reflectometry (TDR) reading, or other physical properties that indicate a performance of the system.

9. The computer-implemented method of claim 1, wherein the performance metrics are software metrics which can include:

a system throughput, a transaction latency, a queue length, a load on a central processing unit, a load on a memory, a load on a cache, I/O traffic, a bus saturation metric, First-In, First-Out (FIFO) overflow statistics, or other software metrics that indicate a performance of the system.

10. An apparatus for determining an estimated value for a sensor which has been removed from a system, comprising:

a sampling mechanism configured to sample values for performance metrics using external sensors that are coupled to a system and internal sensors that are built into the system, wherein at least one internal sensor is a software sensor;

a generating mechanism configured to generate an inferential sensing model for the system from the sampled values; and

an estimating mechanism;

wherein for one or more external sensors that have been removed from the system;

the sampling mechanism is further configured to sample values of performance metrics using the internal sensors, and

the estimating mechanism is configured to generate estimated values for performance metrics for the one or more removed external sensors by using the sampled values from the internal sensors as inputs to the inferential sensing model.

11. The apparatus of claim 10, wherein during operation, the sampling mechanism is configured to sample values of performance metrics using the internal sensors and the estimating mechanism is configured to use the inferential sensing model to generate estimated values for performance metrics for a removed internal sensor from the sampled values from a set of internal sensors which does not include the removed internal sensor.

12. The apparatus of claim 10, wherein during operation, the sampling mechanism is configured to sample values of performance metrics using the internal sensors and the estimating mechanism is configured to use the inferential sensing model to generate estimated values for performance metrics for a first internal sensor from the sampled values from a set of internal sensors which does not include the first internal sensor.

13. The apparatus of claim 10, wherein the apparatus further comprises a warning mechanism configured to determine if the sampled values or the estimated values for the performance metrics have exceeded a threshold and, if so, to generate an alarm.

14. The apparatus of claim 10, wherein when sampling the performance metrics, the sampling mechanism is configured to:

execute one or more test loads on the system; and

sample the performance metrics while the test loads are executing on the system.

15. The apparatus of claim 10, wherein when generating an inferential sensing model, the generating mechanism is configured to use a Logical Combination of Regression Primitives (LCORP) technique, a Non-Linear Non-Parametric (NLNP) regression technique, or another statistical or regression technique to generate a mathematical model that correlates sampled values of performance metrics.

16. The apparatus of claim 10, wherein the system is:

a chip;

a component;

a field-replaceable unit (FRU);

or

a computer system.

17. The apparatus of claim 10, wherein the performance metrics are physical system properties which can include:

a relative humidity, a cumulative or differential vibration, a fan speed, an acoustic signal, a current, a voltage, a time-domain reflectometry (TDR) reading, or other physical properties that indicate a performance of the system.

18. The apparatus of claim 10, wherein the performance metrics are software metrics which can include:

a system throughput, a transaction latency, a queue length, a load on a central processing unit, a load on a memory, a load on a cache, I/O traffic, a bus saturation metric, First-In, First-Out (FIFO) overflow statistics, or other software metrics that indicate a performance of the system.

19. A computer system for determining an estimated value for a sensor which has been removed from a computer system, comprising:

a processor;

a memory coupled to the processor, wherein the memory is configured to store data and instructions for the processor;

a sampling mechanism configured to sample values for performance metrics using external sensors that are coupled to a system and internal sensors that are built into a computer system, wherein at least one internal sensor is a software sensor;

a generating mechanism configured to generate an inferential sensing model for a computer system from the sampled values; and

an estimating mechanism;

wherein for one or more external sensors that have been removed from a computer system;

the sampling mechanism is further configured to sample values of performance metrics using the internal sensors, and

the estimating mechanism is configured to generate estimated values for performance metrics for the one or more removed external sensors by using the sampled values from the internal sensors as inputs to the inferential sensing model.

20. The computer system of claim 19, wherein during operation, the sampling mechanism is configured to sample values of performance metrics using the internal sensors and the estimating mechanism is configured to use the inferential sensing model to generate estimated values for performance metrics for a removed internal sensor from the sampled values from a set of internal sensors which does not include the removed internal sensor.

21. The computer system of claim 19, wherein during operation, the sampling mechanism is configured to sample values of performance metrics using the internal sensors and the estimating mechanism is configured to use the inferential sensing model to generate estimated values for performance metrics for a first internal sensor from the sampled values from a set of internal sensors which does not include the first internal sensor.