Machine learning based end to end system for tcp optimization

1. A computer-implemented method, comprising:

receiving, by one or more computing devices, a plurality of bypass network traffic records for a web application that communicates with user devices from a plurality of different access networks in a time window over a plurality of application servers located at a plurality of different geographic locations, wherein the plurality of bypass network traffic records is generated from network traffic data devoid of information about (a) network access technologies used by the user devices and (b) locations of the user devices;

wherein the plurality of bypass network traffic records is clustered into a plurality of network performance categories, wherein each network performance category in the plurality of network performance categories comprises a respective subset of bypass network traffic records in the plurality of bypass network traffic records;

calculating, by the one or more computing devices, a plurality of sets of sufficient statistics of one or more network optimization parameters for the plurality of network performance categories, wherein each set of sufficient statistics of the one or more network optimization parameters is calculated for a corresponding network performance category based on its respective subset of bypass network traffic records;

partitioning, by the one or more computing devices, the respective subset of bypass network traffic records for the corresponding network performance category into one or more network traffic buckets, thereby generating a plurality of network traffic buckets for the plurality of network performance categories;

generating, from the plurality of sets of sufficient statistics and the plurality of network traffic buckets, a plurality of network quality mappings;

using the plurality of network quality mappings as training instances to train a machine learner for generating network optimization policies; and

causing the one or more network optimization policies generated by the machine learner to be propagated to one or more user devices to be implemented by the one or more user devices in making network requests to the web application.

2. The method as recited in claim 1 , wherein each network quality mapping in the plurality of network quality mappings is weighed by a respective traffic share represented in a network traffic bucket used to generate the network quality mapping.

3. The method as recited in claim 1 , wherein expected values of the network optimization parameters are generated from the sufficient statistics of the network optimization parameters to be used as optimized values of the network optimization parameters.

4. The method as recited in claim 1 , wherein the sufficient statistics of the network optimization parameters are used to generate optimized values of the network optimization parameters.

5. The method as recited in claim 1 , wherein the plurality of network quality mappings is derived from the network traffic data intrinsically with no dependence on information about (a) network access technologies used by the user devices and (b) locations of the user devices.

6. The method as recited in claim 1 , wherein the plurality of bypass network traffic records for the web application is free of network traffic records for accelerated network requests and accelerated data downloads.

7. The method as recited in claim 1 , wherein the plurality of bypass network traffic records for the web application further includes network traffic records for accelerated network requests and accelerated data downloads.

8. A non-transitory computer readable medium storing a program of instructions that is executable by a device to perform a method, the method comprising:

receiving, by a computing device implementing unsupervised machine learning, a plurality of bypass network traffic records for a web application that communicates with user devices from a plurality of different access networks in a time window over a plurality of application servers located at a plurality of different geographic locations, wherein the plurality of bypass network traffic records is generated from network traffic data devoid of information about (a) network access technologies used by the user devices and (b) locations of the user devices;

wherein the plurality of bypass network traffic records is clustered into a plurality of network performance categories, wherein each network performance category in the plurality of network performance categories comprises a respective subset of bypass network traffic records in the plurality of bypass network traffic records;

calculating, using a sufficient statistics generator implemented by the computing device, a plurality of sets of sufficient statistics of one or more network optimization parameters for the plurality of network performance categories, wherein each set of sufficient statistics of the one or more network optimization parameters is calculated for a corresponding network performance category based on its respective subset of bypass network traffic records;

partitioning, using a data aggregator implemented by the computing device, the respective subset of bypass network traffic records for the corresponding network performance category into one or more network traffic buckets, thereby generating a plurality of network traffic buckets for the plurality of network performance categories;

generating, from the plurality of sets of sufficient statistics and the plurality of network traffic buckets, a plurality of network quality mappings;

using the plurality of network quality mappings as training instances to train a machine learner for generating network optimization policies; and

causing the one or more network optimization policies generated by the machine learner to be propagated to one or more user devices to be implemented by the one or more user devices in making network requests to the web application.

9. The non-transitory computer readable medium as recited in claim 8 , wherein each network quality mapping in the plurality of network quality mappings is weighed by a respective traffic share represented in a network traffic bucket used to generate the network quality mapping.

10. The non-transitory computer readable medium as recited in claim 8 , wherein expected values of the network optimization parameters are generated from the sufficient statistics of the network optimization parameters to be used as optimized values of the network optimization parameters.

11. The non-transitory computer readable medium as recited in claim 8 , wherein the sufficient statistics of the network optimization parameters are used to generate optimized values of the network optimization parameters.

12. The non-transitory computer readable medium as recited in claim 8 , wherein the plurality of network quality mappings is derived from the network traffic data intrinsically with no dependence on information about (a) network access technologies used by the user devices and (b) locations of the user devices.

13. The non-transitory computer readable medium as recited in claim 8 , wherein the plurality of bypass network traffic records for the web application is free of network traffic records for accelerated network requests and accelerated data downloads.

14. The non-transitory computer readable medium as recited in claim 8 , wherein the plurality of bypass network traffic records for the web application further includes network traffic records for accelerated network requests and accelerated data downloads.

15. An apparatus, comprising:

one or more computing devices;

a non-transitory computer readable medium storing a program of instructions that is executable by the one or more computing devices to perform a method, the method comprising:

receiving a plurality of bypass network traffic records for a web application that communicates with user devices from a plurality of different access networks in a time window over a plurality of application servers located at a plurality of different geographic locations, wherein the plurality of bypass network traffic records is generated from network traffic data devoid of information about (a) network access technologies used by the user devices and (b) locations of the user devices;

wherein the plurality of bypass network traffic records is clustered into a plurality of network performance categories, wherein each network performance category in the plurality of network performance categories comprises a respective subset of bypass network traffic records in the plurality of bypass network traffic records;

calculating a plurality of sets of sufficient statistics of one or more network optimization parameters for the plurality of network performance categories, wherein each set of sufficient statistics of the one or more network optimization parameters is calculated for a corresponding network performance category based on its respective subset of bypass network traffic records;

partitioning the respective subset of bypass network traffic records for the corresponding network performance category into one or more network traffic buckets, thereby generating a plurality of network traffic buckets for the plurality of network performance categories;

generating, from the plurality of sets of sufficient statistics and the plurality of network traffic buckets, a plurality of network quality mappings;

using the plurality of network quality mappings as training instances to train a machine learner for generating network optimization policies; and

causing the one or more network optimization policies generated by the machine learner to be propagated to one or more user devices to be implemented by the one or more user devices in making network requests to the web application.

16. The apparatus as recited in claim 15 , wherein each network quality mapping in the plurality of network quality mappings is weighed by a respective traffic share represented in a network traffic bucket used to generate the network quality mapping.

17. The apparatus as recited in claim 15 , wherein expected values of the network optimization parameters are generated from the sufficient statistics of the network optimization parameters to be used as optimized values of the network optimization parameters.

18. The apparatus as recited in claim 15 , wherein the sufficient statistics of the network optimization parameters are used to generate optimized values of the network optimization parameters.

19. The apparatus as recited in claim 15 , wherein the plurality of network quality mappings is derived from the network traffic data intrinsically with no dependence on information about (a) network access technologies used by the user devices and (b) locations of the user devices.

20. The apparatus as recited in claim 15 , wherein the plurality of bypass network traffic records for the web application is free of network traffic records for accelerated network requests and accelerated data downloads.

21. The apparatus as recited in claim 15 , wherein the plurality of bypass network traffic records for the web application further includes network traffic records for accelerated network requests and accelerated data downloads.