Autonomous metric tracking and adjustment

• US 9,143,403 B2
• Filed: 06/19/2013
• Issued: 09/22/2015
• Est. Priority Date: 06/19/2013
• Status: Expired due to Fees

• Current Assignee: Hewlett Packard Enterprise Development LP (Hewlett-Packard Enterprise Company)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Faraboschi, Paolo, Milojicic, Dejan S, Barron, Dwight L
• Primary Examiner(s): Vostal, O. C.
• Application Number:
  US13/921,343
• Publication Number:
  US 20140379889A1
• Time in Patent Office: 825 Days
• US Class Current: 1/1
• CPC Class Codes: H04L 41/0823 characterised by the purposes of a change of settings, e.g. optimising configuration for enhancing reliability for optimising operational conditions of wireless networks H04W24/02
• CPC Class Codes: H04L 41/0896 Bandwidth or capacity management, i.e. automatically increasing or decreasing capacities flow or congestion control using dynamic resource allocation, e.g. in-call renegotiation, H04L47/76
• CPC Class Codes: H04L 43/08 Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters

Abstract

Example embodiments relate to autonomous metric tracking and adjustment. In some examples, a computing node may include a processor to run a main operating system and an application that runs on top of the main operating system. The computing node may include a hardware-level controller that dynamically adjusts individual hardware components of the computing node via control signals that do not pass through the main operating system. The adjustments may be based on a target metric from a scheduling service external to the computing node and individual performance metrics from the computing node.

Claims

1. A computing node capable of autonomous metric tracking and adjustment, the computing node comprising:

a processor to run a main operating system;

an application that runs on top of the main operating system; and

a hardware-level controller that dynamically adjusts individual hardware components of the computing node via control signals that do not pass through the main operating system, wherein the adjustments are based on a target metric from a scheduling service external to the computing node and individual performance metrics from the computing node.

2. The computing node of claim 1, wherein the hardware-level controller is further to:

generate a unified observed metric based on the individual performance metrics; and

cause the unified observed metric to be sent to the scheduling service such that the scheduling service can update the target metric based on the unified observed metric and observed metrics from other computing nodes.

3. The computing node of claim 1, wherein the hardware-level controller is integrated in a management controller of the computing node.

4. The computing node of claim 1, wherein the hardware-level controller executes tasks, including the dynamic adjustment, using a processor of the management controller that is separate from the processor that runs the main operating system.

5. The computing node of claim 1, wherein the application communicates with computing devices external to the computing node via a first network and the hardware-level controller communicates with the scheduling service via a second network.

6. The computing node of claim 1, wherein the hardware-level controller uses a closed control loop to adjust the control signals based on a comparison between the target metric and a unified observed metric that is based on the individual performance metrics, wherein the hardware-level controller adjusts the control signals to converge the unified observed metric to the target metric.

7. The computing node of claim 1, wherein the individual performance metrics include performance metrics from the main operating system.

8. The computing node of claim 7, wherein the performance metrics from the main operating system include direct performance metrics that are sent from the application to the hardware-level controller, wherein the performance metrics sent from the application are generated based on programming calls to the hardware-level controller included in the code of the application.

9. The computing node of claim 7, wherein the performance metrics from the main operating system include indirect performance metrics that approximate a progress or rate of execution of the application, wherein the indirect performance metrics include operating system performance counters.

10. The computing node of claim 1, wherein the individual performance metrics include indirect performance metrics from the individual hardware components that approximate a progress or rate of execution of the application.

11. A method for autonomous metric tracking and adjustment executed in a computing node, the method comprising:

receiving, via a hardware-level controller of the computing node, a target metric from a scheduling service external to the computing node;

receiving, via the hardware-level controller, individual performance metrics from the computing node;

determining, via the hardware-level controller, adjustments to be made to individual hardware components of the computing node based on the target metric and the individual performance metrics; and

dynamically adjusting, via the hardware-level controller, the individual hardware components using control signals that do not pass through a main operating system of the computing node.

12. The method of claim 11, wherein the target metric is a single value that is generated based on multiple target performance goals.

13. The method of claim 12, further comprising receiving, from the scheduling service, an indication regarding how the single-value target metric is to be interpreted.

14. The method of claim 11, further comprising:

generating, via the hardware-level controller, a unified observed metric based on the individual performance metrics; and

causing the unified observed metric to be sent to the scheduling service such that the scheduling service can update the target metric based on the unified observed metric and observed metrics from other computing nodes.

15. The method of claim 14, further comprising:

comparing the unified observed metric to the target metric to determine an error; and

using the error for the determining of the adjustments to be made to the individual hardware components.

16. The method of claim 14, further comprising:

receiving a network synchronized time from a time server;

locally advancing the time until the unified observed metric is ready to be sent to the scheduling service; and

inserting a timestamp based on the time into the unified observed metric before the unified observed metric is sent to the scheduling service, wherein the scheduling service uses the timestamp and timestamps of observed metrics of other computing nodes to synchronize information received from and sent to the computing node and the other computing nodes.

17. A system capable of autonomous metric tracking and adjustment, the system comprising:

a computing node disposed in a rack or chassis with other computing nodes, the computing node having a processor to run a main operating system, and an application that runs on top of the main operating system; and

a chassis management controller that is to adjust individual hardware components of the computing node and the other computing nodes in the rack, wherein the chassis management controller includes, for the computing node, a hardware-level controller that dynamically adjusts individual hardware components of the computing node via control signals that do not pass through the main operating system, wherein the adjustments are based on a target metric from a scheduling service external to the chassis and individual performance metrics from the computing node.

18. The system of claim 17, wherein the hardware-level controller uses a closed control loop to adjust the control signals based on a comparison between the target metric and a unified observed metric that is based on the individual performance metrics, wherein the hardware-level controller adjusts the control signals to converge the unified observed metric to the target metric.