Autonomous warehouse-scale computers

Patent metadata

Field	Value
Publication number	US-11960936-B2
Application number	US-202117150285-A
Country	US
Kind code	B2
Filing date	Jan 15, 2021
Priority date	Jan 15, 2021
Publication date	Apr 16, 2024
Grant date	Apr 16, 2024

Abstract

The subject matter described herein provides systems and techniques to address the challenges of growing hardware and workload heterogeneity using a Warehouse-Scale Computer (WSC) design that improves the efficiency and utilization of WSCs. The WSC design may include an abstraction layer and an efficiency layer in the software stack of the WSC. The abstraction layer and the efficiency layer may be designed to improve job scheduling, simplify resource management, and drive hardware-software co-optimization using machine learning techniques and automation in order to customize the WSC for applications at scale. The abstraction layer may embrace platform/hardware and workload diversity through greater coordination between hardware and higher layers of the WSC software stack in the WSC design. The efficiency layer may employ machine learning techniques at scale to realize hardware/software co-optimizations as a part of the autonomous WSC design.

First claim

The invention claimed is:

1. A method of scheduling a workload to be processed on one or more computer servers, the method comprising:
generating a performance profile for the workload using a first software layer;
classifying the workload using a machine learning model based on the performance profile for the workload using the first software layer;
generating a record of selected platforms for the workload using the first software layer; and
scheduling the workload for processing on a hardware layer based on the record of preferred platforms for the workload using a second software layer, wherein the second software layer is disposed between the first software layer and the hardware layer in a software stack of the one or more computer servers.

2. The method of claim 1 , wherein the one or more computer servers are associated with a warehouse-scale computer (WSC).

3. The method of claim 1 , wherein the first software layer is a warehouse-scale computer (WSC) efficiency layer, and wherein the WSC efficiency layer includes at least one software module.

4. The method of claim 3 , wherein the machine learning model includes a hierarchical agglomerative clustering technique, and wherein the at least one software module includes a platform ranking algorithm module that performs the hierarchical agglomerative clustering technique.

5. The method of claim 4 , wherein the platform ranking algorithm module optimizes an objective function that includes at least one of a cost value, an efficiency value, and a performance value associated with scheduling the workload to be processed.

6. The method of claim 1 , wherein the second software layer is a software-defined server (SDS) abstraction layer, and wherein the SDS abstraction layer comprises application programming interfaces (APIs) to directly access the hardware layer.

7. The method of claim 1 , wherein the record of selected platforms indicates an order of preference for each one of a set of hardware platforms of the hardware layer.

8. A system for scheduling a workload to be processed on one or more computer servers, the system comprising one or more processors configured to:
generate a performance profile for the workload using a first software layer;
classify the workload using a machine learning model based on the performance profile for the workload using the first software layer;
generate a record of selected platforms for the workload using the first software layer; and
schedule the workload for processing on a hardware layer based on the record of preferred platforms for the workload using a second software layer, wherein the second software layer is disposed between the first software layer and the hardware layer in a software stack of the one or more computer servers.

9. The system of claim 8 , wherein the one or more computer servers are associated with a warehouse-scale computer (WSC).

10. The system of claim 8 , wherein the first software layer is a warehouse-scale computer (WSC) efficiency layer, and wherein the WSC efficiency layer includes at least one software module.

11. The system of claim 10 , wherein the machine learning model includes a hierarchical agglomerative clustering technique, and wherein the at least one software module includes a platform ranking algorithm module that performs the hierarchical agglomerative clustering technique.

12. The system of claim 11 , wherein the platform ranking algorithm module optimizes an objective function that includes at least one of a cost value, an efficiency value, and a performance value associated with scheduling the workload to be processed.

13. The system of claim 8 , wherein the second software layer is a software-defined server (SDS) abstraction layer, and wherein the SDS abstraction layer comprises application programming interfaces (APIs) to directly access the hardware layer.

14. The system of claim 8 , wherein the record of selected platforms indicates an order of preference for each one of a set of hardware platforms of the hardware layer.

15. A non-transitory computer-readable medium storing instructions for scheduling a workload to be processed on one or more computer servers, that when executed by one or more processors, cause the one or more processors to:
generate a performance profile for the workload using a first software layer;
classify the workload using a machine learning model based on the performance profile for the workload using the first software layer;
generate a record of selected platforms for the workload using the first software layer; and
schedule the workload for processing on a hardware layer based on the record of preferred platforms for the workload using a second software layer, wherein the second software layer is disposed between the first software layer and the hardware layer in a software stack of the one or more computer servers.

16. The non-transitory computer-readable medium of claim 15 , wherein the one or more computer servers are associated with a warehouse-scale computer (WSC).

17. The non-transitory computer-readable medium of claim 15 , wherein the first software layer is a warehouse-scale computer (WSC) efficiency layer, and wherein the WSC efficiency layer includes at least one software module.

18. The non-transitory computer-readable medium of claim 17 , wherein the machine learning model includes a hierarchical agglomerative clustering technique, and wherein the at least one software module includes a platform ranking algorithm module that performs the hierarchical agglomerative clustering technique.

19. The non-transitory computer-readable medium of claim 18 , wherein the platform ranking algorithm module optimizes an objective function that includes at least one of a cost value, an efficiency value, and a performance value associated with scheduling the workload to be processed.

20. The non-transitory computer-readable medium of claim 15 , wherein the second software layer is a software-defined server (SDS) abstraction layer, and wherein the SDS abstraction layer comprises application programming interfaces (APIs) to directly access the hardware layer.

Assignees

• Google Llc

Inventors

• Lo David
• Cheng Liqun
• Ranganathan Parthasarathy
• Dev Sundar Jayakumar

Classifications

• G06F9/5027 Primary

  the resource being a machine, e.g. CPUs, Servers, Terminals · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F9/4881

  Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues · CPC title