Resource allocation in a cloud computing system based on predictions of workload probability parameters

Patent metadata

Field	Value
Publication number	US-12106147-B2
Application number	US-202117352634-A
Country	US
Kind code	B2
Filing date	Jun 21, 2021
Priority date	Jun 21, 2021
Publication date	Oct 1, 2024
Grant date	Oct 1, 2024

Abstract

Disclosed herein are system, method, and computer program product embodiments for allocating resources based on predictions of workload probability parameters. The method can include collecting a first set of historical workload data generated by operating a first set of one or more applications at a first number of past time instances; predicting probability parameters of a second set of future workload data for operating a second set of one or more applications at a second number of future time instances; and determining future resources allocated to operating the second set of one or more applications for the second number of future time instances, based on allocated current resources, a lower bound of resources to satisfy a quality of service (QoS) for operating the second set of one or more applications, an upper bound of resources to satisfy the QoS, and the predicted probability parameters.

First claim

What is claimed is:

1. A computer-implemented method for operating applications on a computing system, comprising:
collecting, by a monitoring operator of a workload manager operated by the computing system, a first set of historical workload data generated by operating a first set of one or more applications at a first number of past time instances, wherein the first set of historical workload data are collected for a set of computing nodes within the computing system, and wherein the monitoring operator monitors historical resources used by the set of computing nodes to operate the first set of one or more applications at the first number of past time instances;
predicting, by the workload manager, probability parameters including a probability density function of a second set of future workload data for operating a second set of one or more applications at a second number of future time instances by the set of computing nodes; and
determining, by the workload manager, future resources allocated to operating the second set of one or more applications for the second number of future time instances, based on allocated current resources, a lower bound number of computing nodes to satisfy a quality of service (QOS) for operating the second set of one or more applications, an upper bound number of computing nodes to satisfy the QoS, and the predicted probability parameters, wherein the future resources are determined as a solution for reducing a first probability for allocating a third number of computing nodes over the upper bound number of computing nodes and reducing a second probability for allocating a fourth number of computing nodes below the lower bound number of computing nodes based on an equation related to the first probability and the second probability;
scheduling the determined future resources of the computing system for the second number of future time instances; and
operating the second set of one or more applications on the scheduled future resources to generate workload data.

2. The method of claim 1 , wherein the first set of one or more applications include same applications as the second set of one or more applications.

3. The method of claim 1 , wherein the second set of one or more applications is different from the first set of one or more applications, and a first application of the first set of one or more applications is an old version of a second application of the second set of one or more applications.

4. The method of claim 1 , wherein the predicting the probability parameters includes predicting the probability parameters of the second set of future workload data based on recurrent neural networks (RNN).

5. The method of claim 1 , wherein the probability parameters include an average of the second set of future workload data, or a standard deviation of the second set of future workload data.

6. The method of claim 1 , wherein the first set of historical workload data and the second set of future workload data include a CPU usage, a memory utilization, a network bandwidth, a latency, a delay, or a throughput.

7. The method of claim 1 , wherein the first number of past time instances includes 5 past time instances, and the second number of future time instances includes 5 future time instances.

8. The method of claim 1 , wherein the first number of past time instances or the second number of future time instances includes a first time instance and a second time instance separated from the first time instance by 1 hour.

9. The method of claim 1 , wherein the first set of historical workload data and the second set of future workload data follow a Gaussian distribution or a Poisson distribution.

10. The method of claim 1 , wherein the future resources and the current resources include a number of computing nodes of the computing system, a number of storage units of the computing system, or an allocation of network bandwidth.

11. The method of claim 1 , wherein the determining future resources includes determining an offset resource to increase or decrease the current resources.

12. A computing system, comprising:
a storage device configured to store a first set of historical workload data generated by operating one or more applications at a first number of past time instances, wherein the first set of historical workload data are collected by a monitoring operator for a set of computing nodes within the computing system, and wherein the monitoring operator monitors historical resources used by the set of computing nodes to operate the one or more applications at the first number of past time instances;
at least one processor coupled to the storage device;
a workload forecasting operator operated by the at least one processor, configured to predict probability parameters including a probability density function of a second set of future workload data for operating the one or more applications at a second number of future time instances by the set of computing nodes; and
a resource allocation scaler operated by the at least one processor, configured to determine future resources allocated to operating the one or more applications for the second number of future time instances, based on allocated current resources, a lower bound number of computing nodes to satisfy a quality of service (QOS) for operating the one or more applications, an upper bound number of computing nodes to satisfy the QoS, and the predicted probability parameters, wherein the future resources are determined as a solution for reducing a first probability for allocating a third number of computing nodes over the upper bound number of computing nodes and reducing a second probability for allocating a fourth number of computing nodes below the lower bound number of computing nodes based on an equation related to the first probability and the second probability; and
a scheduler operated by the at least one processor, configured to schedule the determined future resources of the computing system for the second number of future time instances; and
a processing operator operated by the at least one processor, configured to operate the second set of one or more applications on the scheduled future resources to generate workload data.

13. The system of claim 12 , wherein the workload forecasting operator is configured to predict the probability parameters of the second set of future workload data based on recurrent neural networks (RNN).

14. The system of claim 12 , wherein the probability parameters include an average of the second set of future workload data, or a standard deviation of the second set of future workload data.

15. The system of claim 12 , wherein the first set of historical workload data and the second set of future workload data include a CPU usage, a memory utilization, a network bandwidth, a latency, a delay, or a throughput.

16. The system of claim 12 wherein the first set of historical workload data and the second set of future workload data follow a Gaussian distribution or a Poisson distribution.

17. A non-transitory computer-readable medium having instructions stored thereon that, when executed by at least one computing device, cause the at least one computing device to perform operations comprising:
collecting a first set of historical workload data generated by operating one or more applications at a first number of past time instances, wherein the first set of historical workload data are collected by a monitoring operator for a set of computing nodes within a computing system, and wherein the monitoring operator monitors historical resources used by the set of computi

Assignees

• Sap Se

Inventors

• Guo Jing
• Yang Zhou
• Yan Ming
• Wang Gaozhi
• Cheng Shifan
• Zeng Zhaoxing
• Li Jingtao

Classifications

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N3/08

  Learning methods · CPC title

• G06N3/047

  Probabilistic or stochastic networks · CPC title