Software-based power management for virtualization platforms
A predictive modeling approach to managing processing unit states is disclosed. In an embodiment, a method includes receiving data associated with a host device, including processor statistics. The method generates a feature vector for the host device which includes data derived from the processor statistics and classifies the feature vector using a predictive model. The predictive model can generate a processor state for the host device based on the feature vector. In response, the method can then issue a processor state change message to the host device based on the processor state. The processor state change message causes the host device to reduce power consumption of one or more processing units based on the processor state.
1. A method comprising:
receiving data associated with a host device, the data comprising processor statistics of the host device;
generating a feature vector for the host device, the feature vector including data derived from the processor statistics;
classifying the feature vector using a predictive model, the predictive model generating a processor state for the host device based on the feature vector; and
issuing a processor state change message to the host device based on the processor state, the processor state change message causing the host device to reduce power consumption of one or more processing units based on the processor state.
2. The method of claim 1 , wherein the processor statistics of the host device comprise one or more of a total number of processing units of the host device, a total number of active processing units of the host device, and a total number of inactive processing units of the host device.
3. The method of claim 1 , wherein generating a feature vector for the host device comprises combining the data derived from the processor statistics with static data associated with the host device.
4. The method of claim 1 , wherein classifying the feature vector using a predictive model comprises inputting the feature vector into one of a decision tree, random forest, or naïve Bayes classifier.
5. The method of claim 1 , wherein generating a processor state comprises generating a processor state value, the processor state value comprising one of a c-state value or p-state value.
6. The method of claim 5 , wherein generating a processor state comprises generating at least one processing unit identifier associated with the processor state value.
7. The method of claim 1 , wherein issuing a processor state change message to the host device based on the processor state comprises issuing the processor state change message to the host device using a secure channel, the host device configured to execute a script or binary to change a processor state of the host device.
8. A non-transitory computer-readable storage medium for tangibly storing computer program instructions capable of being executed by a computer processor, the computer program instructions defining steps of:
receiving data associated with a host device, the data comprising processor statistics of the host device;
generating a feature vector for the host device, the feature vector including data derived from the processor statistics;
classifying the feature vector using a predictive model, the predictive model generating a processor state for the host device based on the feature vector; and
issuing a processor state change message to the host device based on the processor state, the processor state change message causing the host device to reduce power consumption of one or more processing units based on the processor state.
9. The non-transitory computer-readable storage medium of claim 8 , wherein the processor statistics of the host device comprise one or more of a total number of processing units of the host device, a total number of active processing units of the host device, and a total number of inactive processing units of the host device.
10. The non-transitory computer-readable storage medium of claim 8 , wherein generating a feature vector for the host device comprises combining the data derived from the processor statistics with static data associated with the host device.
11. The non-transitory computer-readable storage medium of claim 8 , wherein classifying the feature vector using a predictive model comprises inputting the feature vector into one of a decision tree, random forest, or naïve Bayes classifier.
12. The non-transitory computer-readable storage medium of claim 8 , wherein generating a processor state comprises generating a processor state value, the processor state value comprising one of a c-state value or p-state value.
13. The non-transitory computer-readable storage medium of claim 12 , wherein generating a processor state comprises generating at least one processing unit identifier associated with the processor state value.
14. The non-transitory computer-readable storage medium of claim 8 , wherein issuing a processor state change message to the host device based on the processor state comprises issuing the processor state change message to the host device using a secure channel, the host device configured to execute a script or binary to change a processor state of the host device.
15. A device comprising:
a processor configured to:
receive data associated with a host device, the data comprising processor statistics of the host device;
generate a feature vector for the host device, the feature vector including data derived from the processor statistics;
classify the feature vector using a predictive model, the predictive model generating a processor state for the host device based on the feature vector; and
issue a processor state change message to the host device based on the processor state, the processor state change message causing the host device to reduce power consumption of one or more processing units based on the processor state.
16. The device of claim 15 , wherein the processor statistics of the host device comprise one or more of a total number of processing units of the host device, a total number of active processing units of the host device, and a total number of inactive processing units of the host device.
17. The device of claim 15 , wherein generating a feature vector for the host device comprises combining the data derived from the processor statistics with static data associated with the host device.
18. The device of claim 15 , wherein classifying the feature vector using a predictive model comprises inputting the feature vector into one of a decision tree, random forest, or naïve Bayes classifier.
19. The device of claim 15 , wherein generating a processor state comprises generating a processor state value, the processor state value comprising one of a c-state value or p-state value.
20. The device of claim 15 , wherein issuing a processor state change message to the host device based on the processor state comprises issuing the processor state change message to the host device using a secure channel, the host device configured to execute a script or binary to change a processor state of the host device.