Dynamic prediction of system resource requirement of network software in a live network using data driven models

1. A method of automating model creation for a network device, the method comprising:

conducting, by one or more processors and based on a plurality of configuration parameters for the network device, each of a plurality of simulation iterations within a test environment to collect a corresponding plurality of simulation datasets and corresponding operational data representative of an operating state of the network device;

performing, by the one or more processors, a regression analysis with respect to each of the plurality of configuration parameters and each of the plurality of simulation datasets to generate a light weight model representative of the network device that predicts, responsive to the configuration parameters for the network device, the operating state of the network device when configured with the configuration parameters for the network device, the light weight model configured to consume, during execution of the light weight model, less computing resources compared to execution of a model generated using machine learning with respect to the configuration parameters and the corresponding operational data; and

outputting, by the one or more processors, the light weight model for installation on a computing resource restricted network device to enable prediction of an operating state of the computing resource restricted network device relative to changes to the configuration parameters prior to committing the changes to the configuration parameters and committing the changes to the configuration parameters when the changes to the configuration parameters are not predicted to result in the operating state of the computing resource restricted network device exceeding a threshold operating state,

wherein the computing resource restricted network device comprises a router having a control plane and a forwarding plane,

wherein the forwarding plane includes a packet forwarding engine,

wherein the operational data include resource utilization data indicative of resource utilization within the network device, and

wherein the resource utilization data comprises one or more of resource usage per control plane application executed by the control plane of the network device during each simulation, and resource usage in the packet forwarding engine of the network device during each simulation.

2. The method of claim 1 , wherein performing the regression analysis comprises:

performing the regression analysis to identify a relationship, expressed as one or more coefficients, between the configuration parameters for the network device and the operating state of the network device; and

configuring a regression model using the one or more coefficients to obtain the light weight model.

3. The method of claim 2 ,

wherein performing the regression analysis comprises performing a polynomial regression analysis to identify the relationship, expressed as one or more polynomial coefficients and one or more intercepts, between the configuration parameters for the network device and the operating state of the network device, and

wherein configuring the regression model comprises configuring a polynomial regression model using the one or more polynomial coefficients and the one or more intercepts to obtain the light weight model.

4. The method of claim 2 , wherein performing the regression analysis comprises performing the regression analysis to identify the relationship as a fit between the configuration parameters for the network device and the operating state of the network device within a threshold accuracy to ensure a stated accuracy.

5. The method of claim 3 , wherein the polynomial regression model includes polynomial basis functions used to define the relationship as a non-linear relationship, expressed as the one or more polynomial coefficients and the one or more intercepts, between the configuration parameters for the network device and the operating state of the network device.

6. The method of claim 3 , wherein the one or more polynomial coefficients include at most two polynomial coefficients to express the relationship between the configuration parameters for the network device and the operating state of the network device.

7. A device configured to automate model creation for a network device, the device comprising:

one or more processors configured to:

conduct, based on a plurality of configuration parameters, each of a plurality of simulation iterations within a test environment to collect a corresponding plurality of simulation datasets that each include operational data representative of operating states of the network device;

perform a regression analysis with respect to each of the plurality of configuration parameters and each of the plurality of simulation datasets to generate a light weight model representative of the network device that predicts, responsive to particular configuration parameters for the network device, an operating state of the network device when configured with the particular configuration parameters for the network device, the light weight model configured to consume, during execution of the light weight model, less computing resources compared to execution of a model generated using machine learning with respect to the plurality of configuration parameters and each of the plurality of simulation datasets; and

output the light weight model for installation on a computing resource restricted network device to enable prediction of an operating state of the computing resource restricted network device relative to changes to the configuration parameters prior to committing the changes to the configuration parameters and committing the changes to the configuration parameters when the changes to the configuration parameters are not predicted to result in the operating state of the computing resource restricted network device exceeding a threshold operating state,

wherein the computing resource restricted network device comprises a router having a control plane and a forwarding plane,

wherein the forwarding plane includes a packet forwarding engine,

wherein the operational data include resource utilization data indicative of resource utilization within the network device, and

wherein the resource utilization data comprises one or more of resource usage per control plane application executed by the control plane of the network device during each simulation, and resource usage in the packet forwarding engine of the network device during each simulation; and

a memory configured to store the light weight model.

8. The device of claim 7 , wherein the one or more processors are, when configured to perform the regression analysis, configured to:

perform the regression analysis to identify a relationship, expressed as one or more coefficients, between the configuration parameters for the network device and the operating state of the network device; and

configure a regression model using the one or more coefficients to obtain the light weight model.

9. The device of claim 8 ,

wherein the one or more processors are, when configured to perform the regression analysis, configured to perform a polynomial regression analysis to identify the relationship, expressed as one or more polynomial coefficients and one or more intercepts, between the configuration parameters for the network device and the operating state of the network device, and

wherein the one or more processors are, when configured to configure the regression model, configured to configure a polynomial regression model using the one or more polynomial coefficients and the one or more intercepts to obtain the light weight model.

10. The device of claim 8 , wherein the one or more processors are, when configured to perform the regression analysis, configured to perform the regression analysis to identify the relationship as a fit between the configuration parameters for the network device and the operating state of the network device within a threshold accuracy to ensure a stated accuracy.

11. The device of claim 9 , wherein the polynomial regression model includes polynomial basis functions used to define the relationship as a non-linear relationship, expressed as the one or more polynomial coefficients and the one or more intercepts, between the configuration parameters for the network device and the operating state of the network device.

12. The device of claim 9 , wherein the one or more polynomial coefficients include at most two polynomial coefficients to express the relationship between the configuration parameters for the network device and the operating state of the network device.

13. A non-transitory computer-readable storage medium having stored thereon instructions that, when executed to perform model creation for a network device, cause one or more processors to:

conduct, based on a plurality of configuration parameters, each of a plurality of simulation iterations within a test environment to collect a corresponding plurality of simulation datasets that each include operational data representative of operating states of the network device;

perform a regression analysis with respect to each of the plurality of configuration parameters and each of the plurality of simulation datasets to generate a light weight model representative of the network device that predicts, responsive to particular configuration parameters for the network device, an operating state of the network device when configured with the particular configuration parameters for the network device, the light weight model configured to consume, during execution of the light weight model, less computing resources compared to execution of a model generated using machine learning with respect to the plurality of configuration parameters and each of the plurality of simulation datasets; and

output the light weight model for installation on a computing resource restricted network device to enable prediction of an operating state of the computing resource restricted network device relative to changes to the configuration parameters prior to committing the changes to the configuration parameters and committing the changes to the configuration parameters when the changes to the configuration parameters are not predicted to result in the operating state of the computing resource restricted network device exceeding a threshold operating state,

wherein the computing resource restricted network device comprises a router having a control plane and a forwarding plane,

wherein the forwarding plane includes a packet forwarding engine,

wherein the operational data include resource utilization data indicative of resource utilization within the network device, and

wherein the resource utilization data comprises one or more of resource usage per control plane application executed by the control plane of the network device during each simulation, and resource usage in the packet forwarding engine of the network device during each simulation.

14. A system comprising:

a computing device comprising:

one or more processors configured to:

conduct, based on a plurality of configuration parameters, each of a plurality of simulation iterations within a test environment to collect a corresponding plurality of simulation datasets that each include operational data representative of operating states of a first network device;

perform a regression analysis with respect to each of the plurality of configuration parameters and each of the plurality of simulation datasets to generate a light weight model representative of the first network device that predicts, responsive to particular configuration parameters for the first network device, an operating state of the first network device when configured with the particular configuration parameters for the first network device, the light weight model configured to consume, during execution of the light weight model, less computing resources compared to execution of a model generated using machine learning with respect to the plurality of configuration parameters and each of the plurality of simulation datasets; and

output the light weight model for installation on a computing resource restricted network device to enable prediction of an operating state of the computing resource restricted network device relative to changes to the configuration parameters prior to committing the changes to the configuration parameters and committing the changes to the configuration parameters when the changes to the configuration parameters are not predicted to result in the operating state of the computing resource restricted network device exceeding a threshold operating state,

wherein the computing resource restricted network device comprises a router having a control plane and a forwarding plane,

wherein the forwarding plane includes a packet forwarding engine,

wherein the operational data include resource utilization data indicative of resource utilization within the network device, and

wherein the resource utilization data comprises one or more of resource usage per control plane application executed by the control plane of the network device during each simulation, and resource usage in the packet forwarding engine of the network device during each simulation; and

a memory configured to store the light weight model; and

a second network device deployed in an operating environment in support of delivery of one or more services including forwarding of network packets, the second network device comprising:

a hardware interface configured to receive the particular configuration parameters; and

one or more processors configured to:

execute the light weight model to obtain, based on the particular configuration parameters, a predicted operating state of the second network device;

determine if the predicted operating state exceeds one or more operating state thresholds; and

commit the particular configuration parameters only if the predicted operating state does not exceed the one or more operating state thresholds.