Forecasting computer resources demand

• US 10,361,924 B2
• Filed: 04/04/2014
• Issued: 07/23/2019
• Est. Priority Date: 04/04/2014
• Status: Active Grant

• Current Assignee: Kyndryl Incorporated (Kyndryl Holdings, Inc.)
• Original Assignee: International Business Machines Corporation
• Inventors: Baughman, Aaron K., Bogdany, Richard J., Locke, Richard A., O'Connell, Brian M.
• Primary Examiner(s): Ochoa, Juan C
• Application Number:
  US14/245,640
• Publication Number:
  US 20150288571A1
• Time in Patent Office: 1,936 Days
• Field of Search: 703 13
• CPC Class Codes: H04L 41/145 involving simulating, designing, planning or modelling of a network
• CPC Class Codes: H04L 41/147 for predicting network behaviour
• CPC Class Codes: H04L 41/16 using machine learning or artificial intelligence

Abstract

An approach for forecasting demand. The approach includes a method that includes receiving one or more variables associated with an event. The method further includes generating, by at least one computing device, a model to forecast future demand based on the one or more variables. The method further includes determining, by the at least one computing device, a load to provision one or more servers to meet the future demand. The load is based on the model.

Claims

1. A method comprising:

receiving one or more variables associated with an event;

generating, by at least one computing device, a model to forecast future demand based on the one or more variables;

determining, by the at least one computing device, a load to provision one or more cloud servers to meet the future demand, wherein the load is based on the model;

physically provisioning the one or more cloud servers with the determined load based on a trained model which includes a forecasted further demand with a threshold level of error to increase forecasting accuracy during the event; and

automatically allocating additional cloud computing resources to meet the forecasted future demand ahead of a time of the event to prevent an unsustainable volume of web traffic,

wherein the one or more variables are associated with the web traffic associated with at least one of proximity to lead or playoff predictions,

the proximity to lead is based on projecting the web traffic associated with particular participants being within a threshold level of being in first place in the event, and

the model is trained so that the future demand is forecasted with the threshold level of error, and the training of the model includes generating a burst event based on a historical training set and an online training set, and the burst event is used to compute a demand factor feature vector which is used to train the model using a number of coefficients equal to the magnitude value of the demand factor feature vector.

2. The method of claim 1, wherein an error rate associated with the model forecasting the future demand is reduced by training the model with different values associated with the one or more variables.

3. The method of claim 2, wherein the training includes determining a pattern associated with the forecasting the future demand.

4. The method of claim 1, wherein the future demand is associated with cyclic demand and demand spikes over a period of time, and the load is based on at least one of memory, disk space, and processor units.

5. The method of claim 1, wherein the one or more variables are associated with the web traffic further associated with:

web exclusive demand;

participant popularity; and

featured groups.

6. The method of claim 5, wherein the web traffic associated with participant popularity is based on obtaining data from one or more social networking websites which mention a participant'"'"'s name.

7. The method of claim 5, wherein the web traffic associated with the playoff predictions is based on projecting web traffic for a predicted playoff based on analyzing different participants and their scoring capabilities within the event.

8. The method of claim 5, wherein the web traffic associated with the web exclusive demand is based on projected web traffic for content that is only available on the Internet.

9. The method of claim 5, wherein the featured groups is based on projected web traffic associated with participants who are provided web accessible streaming coverage during the event.

10. The method of claim 1, wherein a service provider at least one of creates, maintains, deploys and supports a computer infrastructure which comprises the at least one computing device.

11. The method of claim 1, wherein spikes of demand associated with the future demand are determined by using online information.

12. The method of claim 1, wherein steps of claim 1 are provided by a service provider on a subscription, advertising, and/or fee basis.

13. The method of claim 11, wherein the spikes of demand are bursts of demand that occur intermittingly over a period of time.

14. The method of claim 13, further comprising adding overlap to a forecast curve based on at least one of a forecast of a maximum value, a forecast of a minimum value, and a forecast of an average value, and the overlap being a difference between the spikes of demand based on current information and historical information.

15. The method of claim 1, further comprising:

generating at least one event in a future timeline;

generating at least one predicted burst event based on a historical timeline of a previous event which is similar to the at least one event;

generating a predicted spike in demand based on the at least one predicted burst event using the historical timeline;

generating a spike in demand based on a current timeline;

adding overlap to a forecast curve which is a difference between the predicted spike in demand based on historical information and the spike in demand based on the current timeline;

generating a spike adjusted forecast based on the generated predicted spike in demand, the generated spike in demand, and the overlap; and

determining a load based on the generated spike adjusted forecast.

16. A computer program product for predicting demand in a cloud environment, the computer program product comprising a computer usable storage medium having program code embodied in the storage medium, the program code being readable/executable by a computing device to:

receive one or more variables associated with an event;

generate a model to forecast future demand based on the one or more variables being associated with spikes of demand over a period of time;

determine a load based on the model, wherein the load is used to provision one or more cloud servers to meet the future demand;

physically provision the one or more cloud servers with the determined load based on a trained model which includes a burst event to increase forecasting accuracy during the event; and

automatically allocate additional cloud computing resources to meet the future demand ahead of a time of the event to prevent an unsustainable volume of web traffic,

wherein the model is trained so that the future demand is forecasted with a threshold level of error, and

the training of the model includes generating the burst event based on a historical training set and an online training set, and the burst event is used to compute a demand factor feature vector which is used to train the model using a number of coefficients with a regularized stochastic gradient descent, the number of coefficients being equal to the magnitude value of the demand factor feature vector.

17. The computer program product of claim 16, wherein the model is based on a vector that incorporates data demand predictions for the one or more variables.

18. The computer program product of claim 16, wherein the spikes of demand are adjusted by implementing a forecast curve based on at least one of a forecast of a maximum value, a forecast of a minimum value, and a forecast of an average value.

19. The computer program product of claim 18, wherein the spikes of demand use historical and online periods of information.

20. A system comprising:

a CPU, a computer readable memory and a computer readable storage medium;

program instructions to receive one or more variables associated with an event, wherein the one or variables are associated with data demand for playoff predictions;

program instructions to generate a model to forecast future demand based on the one or more variables being associated with spikes of demand over a period of time, wherein the model uses a demand factor feature vector; and

program instructions to determine a load based on the model, wherein the load is used to provision one or more cloud servers to meet the future demand,

wherein the program instructions are stored on the computer readable storage medium for execution by the CPU via the computer readable memory,

a web traffic associated with the playoff predictions is based on projecting web traffic for a predicted playoff based on analyzing different participants and their scoring capabilities within the event,

physically provision the one or more cloud servers with the determined load based on the generated model which includes a forecasted future demand to increase forecasting accuracy during the event,

automatically allocate additional cloud computing resources to meet the future demand ahead of a time of the event to prevent an unsustainable volume of the web traffic, and

the additional cloud computing resources comprise the one or more cloud servers, at least one processor, and at least one memory device.