Proactively accomodating predicted future serverless workloads using a machine learning prediction model

1. A computer-implemented method comprising:

receiving information indicative of a target value for a performance metric for an application running in a serverless framework of a private cloud;

training a machine-learning prediction model to forecast future serverless workloads during a window of time for the application based on historical serverless workload information associated with the application by performing a training process; and

during the window of time, obtaining a plurality of serverless workload observations for the application from the serverless framework;

predicting a future serverless workload for the application at a future time by applying the machine learning prediction model to the plurality of workload observations;

using a feedback control system that continuously monitors the performance metric to output a suggested number of replicas based on a current value of the performance metric obtained from the serverless framework, the target value, and the predicted future serverless workload, wherein the feedback control system comprises a proportional-integral-derivative (PID) controller; and

prior to the future time, proactively accommodating the predicted future serverless workload by causing the serverless framework to scale a number of replicas supporting the application to the suggested number and pre-warm the replicas.

2. The computer-implemented method of claim 1 , further comprising tuning a plurality of parameters of the PID controller using the Ziegler-Nichols method based on a previous number of replicas supporting the application, the target value, and a reference value representing a value below which a given percentage of observations of the target value fails.

3. The computer-implemented method of claim 1 , wherein the target performance metric comprises a quality of service (QoS) performance requirement.

4. The computer-implemented method of claim 1 , wherein the target performance metric comprises a service level agreement (SLA) metric.

5. The computer-implemented method of claim 1 , wherein the serverless workload is represented by a number of requests per second.

6. The computer-implemented method of claim 1 , wherein the replicas comprise containers.

7. The computer-implemented method of claim 1 , wherein the machine-learning prediction model comprises a support vector regression (SVR) model.

8. The computer-implemented method of claim 7 , wherein said training the machine-learning prediction model to forecast future serverless workloads involves the use of a linear kernel, a polynomial kernel, a radial basis function (RBF) kernel or a sigmoid kernel.

9. The computer-implemented method of claim 7 , wherein the window of time is dynamic, wherein W represents a number of time units in the dynamic window of time, wherein t represents a current time, the method further comprising:

when a workload variability measure involving a maximum observed workload value and a minimum observed workload value of a plurality of serverless workload observations for the application collected during the dynamic window of time is greater than or equal to a first threshold, then decreasing W; and

when the workload variability measure is less than or equal to a second threshold, then increasing W; and

responsive to expiration of the dynamic window of time, retraining the machine-learning prediction model for a new window of time beginning at t−W and ending at t based on the plurality of serverless workload observations for the application collected during the new window of time by performing the training procedure.

10. A non-transitory machine readable medium storing instructions executable by a processing resource of a computer system, the non-transitory machine readable medium comprising instructions to:

receive information indicative of a target value for a performance metric for an application running in a serverless framework of a private cloud;

train a machine-learning prediction model to forecast future serverless workloads during a window of time for the application based on historical serverless workload information associated with the application by performing a training process; and

during the window of time, obtain a plurality of serverless workload observations for the application from the serverless framework;

predict a future serverless workload for the application at a future time by applying the machine learning prediction model to the plurality of workload observations;

use a feedback control system that continuously monitors the performance metric to output a suggested number of replicas based on a current value of the performance metric obtained from the serverless framework, the target value, and the predicted future serverless workload, wherein the feedback control system comprises a proportional-integral-derivative (PID) controller; and

prior to the future time, proactively accommodate the predicted future serverless workload by causing the serverless framework to scale a number of replicas supporting the application to the suggested number and pre-warm the replicas.

11. The non-transitory machine readable medium of claim 10 , wherein the instructions are further to tune a plurality of parameters of the PID controller using the Ziegler-Nichols method based on a previous number of replicas supporting the application, the target value, and a reference value representing a value below which a given percentage of observations of the target value fails.

12. The non-transitory machine readable medium of claim 10 , wherein the target performance metric comprises a quality of service (QoS) performance requirement.

13. The non-transitory machine readable medium of claim 10 , wherein the target performance metric comprises a service level agreement (SLA) metric.

14. The non-transitory machine readable medium of claim 10 , wherein the serverless workload is represented by a number of requests per second.

15. The non-transitory machine readable medium of claim 10 , wherein the replicas comprise containers.

16. The non-transitory machine readable medium of claim 10 , wherein the machine-learning prediction model comprises a support vector regression (SVR) model.

17. The non-transitory machine readable medium of claim 16 , wherein the instructions to train the machine-learning prediction model to forecast future serverless workloads use a linear kernel, a polynomial kernel, a radial basis function (RBF) kernel or a sigmoid kernel.

18. The non-transitory machine readable medium of claim 16 , wherein the window of time is dynamic, wherein W represents a number of time units in the dynamic window of time, wherein t represents a current time and wherein the instructions are further to:

decrease W when a workload variability measure involving a maximum observed workload value and a minimum observed workload value of a plurality of serverless workload observations for the application collected during the dynamic window of time is greater than or equal to a first threshold; and

increase W when the workload variability measure is less than or equal to a second threshold; and

responsive to expiration of the dynamic window of time, retrain the machine-learning prediction model for a new window of time beginning at t−W and ending at t based on the plurality of serverless workload observations for the application collected during the new window of time by performing the training procedure.