End-to-end quality of service in edge computing environments
Systems and techniques for end-to-end quality of service in edge computing environments are described herein. A set of telemetry measurements may be obtained for an ongoing dataflow between a device and a node of an edge computing system. A current key performance indicator (KPI) may be calculated for the ongoing dataflow. The current KPI may be compared to a target KPI to determine an urgency value. A set of resource quality metrics may be collected for resources of the network. The set of resource quality metrics may be evaluated with a resource adjustment model to determine available resource adjustments. A resource adjustment may be selected from the available resource adjustments based on an expected minimization of the urgency value. Delivery of the ongoing dataflow may be modified using the selected resource adjustment.
1. A method for end-to-end quality dataflow management in a network of an edge computing system, comprising:
collecting delivery performance telemetry data for an ongoing quality dataflow between a first node and a second node of the edge computing system;
calculating a current performance metric for the ongoing quality dataflow using the delivery performance telemetry data;
comparing the current performance metric to a target performance metric to determine an urgency value for the ongoing quality dataflow;
evaluating the ongoing quality dataflow for dataflow delivery resources of the network with a resource adjustment model to determine a cost of available resource adjustments; and
modifying delivery of the ongoing quality dataflow using a resource adjustment, the resource adjustment selected based on an expected minimization of cost of delivery of the ongoing quality dataflow and the urgency value for the ongoing quality dataflow.
2. The method of claim 1 , further comprising:
calculating an updated urgency value for the ongoing quality dataflow subsequent to modifying the delivery of the ongoing quality dataflow; and
retraining the resource adjustment model using the resource adjustment and an actual reduction in the cost of delivery of the ongoing quality dataflow and the updated urgency value.
3. The method of claim 1 , wherein the urgency value is determined by applying a cost function to the current performance metric and the target performance metric.
4. The method of claim 1 , further comprising:
generating a quality of service (QoS) prediction model using historical performance metrics for quality data flows; and
predicting a QoS for the ongoing quality dataflow by evaluating the delivery performance telemetry data using the QoS prediction model, wherein the target performance metric is determined using the QoS for the ongoing quality dataflow.
5. The method of claim 1 , further comprising:
generating a demand prediction model using historical telemetry data for the edge computing system; and
calculating a demand prediction by evaluating current telemetry data for the edge computing system using the demand prediction model, wherein the cost of the available resource adjustments is determined in part using the demand prediction.
6. The method of claim 1 , wherein determining the urgency value for the ongoing quality dataflow further comprises:
generating a matrix model using the current performance metric and the target performance metric; and
calculating the urgency value as gap between the current performance metric and the target performance metric in the matrix model, wherein the current performance metric indicates current quality of the ongoing quality dataflow and the target performance metric indicates predicted quality of the ongoing quality dataflow.
7. The method of claim 1 , wherein the ongoing quality dataflow is a quality metrics dataflow.
8. The method of claim 1 , wherein the delivery performance telemetry data includes at least one of a latency value, a throughput value, a frame rate, or a bit rate.
9. A system for end-to-end quality dataflow management in a network of an edge computing system, comprising:
at least one processor; and
memory including instructions that, when executed by the at least one processor, cause the at least one processor to perform operations to:
collect delivery performance telemetry data for an ongoing quality dataflow between a first node and a second node of the edge computing system;
calculate a current performance metric for the ongoing quality dataflow using the delivery performance telemetry data;
compare the current performance metric to a target performance metric to determine an urgency value for the ongoing quality dataflow;
evaluate the ongoing quality dataflow for dataflow delivery resources of the network with a resource adjustment model to determine a cost of available resource adjustments; and
modify delivery of the ongoing quality dataflow using a resource adjustment, the resource adjustment selected based on an expected minimization of cost of delivery of the ongoing quality dataflow and the urgency value for the ongoing quality dataflow.
10. The system of claim 9 , the memory further comprising instructions that, when executed by the at least one processor, cause the at least one processor to perform operations to:
calculate an updated urgency value for the ongoing quality dataflow subsequent to modifying the delivery of the ongoing quality dataflow; and
retrain the resource adjustment model using the resource adjustment and an actual reduction in the cost of delivery of the ongoing quality dataflow and the updated urgency value.
11. The system of claim 9 , wherein the urgency value is determined by application of a cost function to the current performance metric and the target performance metric.
12. The system of claim 9 , the memory further comprising instructions that, when executed by the at least one processor, cause the at least one processor to perform operations to:
generate a quality of service (QoS) prediction model using historical performance metrics for quality data flows; and
predict a QoS for the ongoing quality dataflow by evaluation of the delivery performance telemetry data using the QoS prediction model, wherein the target performance metric is determined using the QoS for the ongoing quality dataflow.
13. The system of claim 9 , the memory further comprising instructions that, when executed by the at least one processor, cause the at least one processor to perform operations to:
generate a demand prediction model using historical telemetry data for the edge computing system; and
calculate a demand prediction by evaluating current telemetry data for the edge computing system using the demand prediction model, wherein the cost of the available resource adjustments is determined in part using the demand prediction.
14. The system of claim 9 , wherein the instructions to determine the urgency value for the ongoing quality dataflow further comprises instructions that, when executed by the at least one processor, cause the at least one processor to perform operations to:
generate a matrix model using the current performance metric and the target performance metric; and
calculate the urgency value as gap between the current performance metric and the target performance metric in the matrix model, wherein the current performance metric indicates current quality of the ongoing quality dataflow and the target performance metric indicates predicted quality of the ongoing quality dataflow.
15. The system of claim 9 , wherein the ongoing quality dataflow is a quality metrics dataflow.
16. The system of claim 9 , wherein the delivery performance telemetry data includes at least one of a latency value, a throughput value, a frame rate, or a bit rate.
17. At least one non-transitory machine-readable medium including instructions for end-to-end quality dataflow management in a network of an edge computing system that, when executed by at least one processor, cause the at least one processor to perform operations to:
collect delivery performance telemetry data for an ongoing quality dataflow between a first node and a second node of the edge computing system;
calculate a current performance metric for the ongoing quality dataflow using the delivery performance telemetry data;
compare the current performance metric to a target performance metric to determine an urgency value for the ongoing quality dataflow;
evaluate the ongoing quality dataflow for dataflow delivery resources of the network with a resource adjustment model to determine a cost of available resource adjustments; and
modify delivery of the ongoing quality dataflow using a resource adjustment, the resource adjustment selected based on an expected minimization of cost of delivery of the ongoing quality dataflow and the urgency value for the ongoing quality dataflow.
18. The at least one non-transitory machine-readable medium of claim 17 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to perform operations to:
calculate an updated urgency value for the ongoing quality dataflow subsequent to modifying the delivery of the ongoing quality dataflow; and
retrain the resource adjustment model using the resource adjustment and an actual reduction in the cost of delivery of the ongoing quality dataflow and the updated urgency value.
19. The at least one non-transitory machine-readable medium of claim 17 , wherein the urgency value is determined by application of a cost function to the current performance metric and the target performance metric.
20. The at least one non-transitory machine-readable medium of claim 17 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to perform operations to:
generate a quality of service (QoS) prediction model using historical performance metrics for quality data flows; and
predict a QoS for the ongoing quality dataflow by evaluation of the delivery performance telemetry data using the QoS prediction model, wherein the target performance metric is determined using the QoS for the ongoing quality dataflow.
21. The at least one non-transitory machine-readable medium of claim 17 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to perform operations to:
generate a demand prediction model using historical telemetry data for the edge computing system; and
calculate a demand prediction by evaluating current telemetry data for the edge computing system using the demand prediction model, wherein the cost of the available resource adjustments is determined in part using the demand prediction.
22. The at least one non-transitory machine-readable medium of claim 17 , wherein the instructions to determine the urgency value for the ongoing quality dataflow further comprises instructions that, when executed by the at least one processor, cause the at least one processor to perform operations to:
generate a matrix model using the current performance metric and the target performance metric; and
calculate the urgency value as gap between the current performance metric and the target performance metric in the matrix model, wherein the current performance metric indicates current quality of the ongoing quality dataflow and the target performance metric indicates predicted quality of the ongoing quality dataflow.
23. The at least one non-transitory machine-readable medium of claim 17 , wherein the ongoing quality dataflow is a quality metrics dataflow.
24. The at least one non-transitory machine-readable medium of claim 17 , wherein the delivery performance telemetry data includes at least one of a latency value, a throughput value, a frame rate, or a bit rate.