Hybrid system-on-chip for power and performance prediction and control
Systems and methods for a hybrid system-on-chip usable for predicting performance and power requirements of a host server include a big cores module, including central processing units, for receiving and pre-processing performance and power metrics data of the host server and to allocate computing resources, a small cores module, including massively parallel processing units, for mapping each instance associated with host server in the performance and power metrics data to a corresponding massively parallel processing unit based on the allocated computing resources for a per-instance metrics calculation, and an artificial intelligence (AI) accelerator for calculating performance and power prediction results based on the per-instance calculations from the small cores module.
1. A device for predicting performance and power (PnP) requirements of a host server, the device comprising:
a big cores module including a plurality of general purpose central processing units (CPUs), the big cores module configured to receive and pre-process PnP metrics data of the host server and to allocate computing resources for calculating the PnP requirements;
a small cores module including a plurality of massively parallel processing (MPP) units, the small cores module coupled to the big cores module, the small cores module configured to map each instance associated with host server in the PnP metrics data to a corresponding MPP unit based on the allocated computing resources for a per-instance metric calculation; and
an artificial intelligence (AI) accelerator coupled to the big cores module and the small cores modules, the AI accelerator configured to calculate PnP prediction results based on the per-instance calculations from the small cores module and to send the PnP prediction results to the big cores module.
2. The device of claim 1 , wherein the small cores module is further configured to perform rule-based prediction, generate rule-based prediction results, and send the rule-based prediction results to big cores module.
3. The device of claim 1 , wherein the big cores module is coupled to the host server via a host interface, the big cores module further configured to send the PnP prediction results to the host server via the host interface.
4. The device of claim 3 , wherein a software agent running on the host server is configured to collect the PnP metrics data to be sent to the big cores module and receive the PnP prediction results from the big cores module.
5. A method for predicting performance and power (PnP) requirements of a host server, the method comprising:
receiving and pre-processing, by a big cores module of a device, PnP metrics data of the host server, the big cores module including a plurality of general purpose central processing units (CPUs);
allocating, by the big cores module, computing resources for calculating the PnP requirements;
mapping, by a small cores module of the device, each instance associated with host server in the PnP metrics data to a corresponding massively parallel processing (MPP) unit based on the allocated computing resources for a per-instance metric calculation, the small cores module including a plurality of MPP units;
calculating, by an artificial intelligence (AI) accelerator of the device, PnP prediction results based on the per-instance calculations from the small cores module; and
sending, by the AI accelerator, the PnP prediction results to the big cores module.
6. The method of claim 5 , further comprising:
performing, by the small cores module, rule-based prediction to generate rule-based prediction results, and
sending, by the small cores module, the rule-based prediction results to the big cores module.
7. The method of claim 5 , further comprising:
sending, by the big cores module, the PnP prediction results to the host server via a host interface.
8. The method of claim 5 , wherein the device is a hybrid system on chip (SoC).
9. The method of claim 7 , further comprising:
collecting, by a software agent running on the host server, the PnP metrics data to be sent to the big cores module; and
receiving, by the software agent, the PnP prediction results from the big cores module via the host interface.
10. The method of claim 9 , wherein:
collecting, by a software agent running on the host server, the PnP metrics data to be sent to the big cores module includes collecting the PnP metrics data at a predetermined collection frequency; and
receiving, by the software agent, the PnP prediction results from the big cores module includes receiving the PnP prediction results from the big cores module at a predetermined receiving frequency.
11. The method of claim 9 , further comprising:
generating, by the software agent, hardware PnP control options and parameters for the hardware PnP control options based on the PnP prediction results.
12. The method of claim 11 , further comprising:
applying, by the software agent, the hardware PnP control options to hardware components of the host server.
13. The method of claim 9 , further comprising:
revising, by the software agent, a scheduler algorithm based on the PnP prediction results.
14. The method of claim 13 , further comprising:
applying, by the software agent, the revised scheduler algorithm to software components of the host server.
15. A non-transitory computer-readable storage medium storing computer-readable instructions executable by one or more processors, that when executed by the one or more processors, cause the one or more processors to perform operations comprising:
collecting performance and power (PnP) metrics data of a host server;
transmitting the PnP metrics data, via a host interface, to a big cores module of a device, the big cores module including a plurality of general purpose central processing units (CPUs);
pre-processing, by the big cores module, the PnP metrics data;
allocating, by the big cores module, computing resources for calculating the PnP requirements;
mapping, by a small cores module of the device, each instance associated with host server in the PnP metrics data to a corresponding massively parallel processing (MPP) unit based on the allocated computing resources for a per-instance metric calculation, the small cores module including a plurality of MPP units;
calculating, by an artificial intelligence (AI) accelerator of the device, PnP prediction results based on the per-instance calculations from the small cores module;
sending, by the AI accelerator, the PnP prediction results to the big cores module;
receiving, by the big cores module, the PnP prediction results from the AI accelerator; sending, by the big cores module, the PnP prediction to the host server via the host interface; and
receiving the PnP prediction results from the big cores module.
16. The non-transitory computer-readable storage medium of claim 15 , wherein the operations further comprise:
generating hardware PnP control options and parameters for the hardware PnP control options based on the PnP prediction results.
17. The non-transitory computer-readable storage medium of claim 16 , wherein the operations further comprise:
applying the hardware PnP control options to hardware components of the host server.
18. The non-transitory computer-readable storage medium of claim 15 , wherein the operations further comprise:
revising a scheduler algorithm based on the PnP prediction results; and
applying the revised scheduler algorithm to software components of the host server.
19. The non-transitory computer-readable storage medium of claim 15 , wherein the operations further comprise:
performing, by the small cores module, rule-based prediction to generate rule-based prediction results, and
sending, by the small cores module, the rule-based prediction results to the big cores module.
20. The non-transitory computer-readable storage medium of claim 15 , wherein the device is a hybrid system-on-chip (SoC).