Recommending configuration changes in software-defined networks using machine learning

An apparatus, comprising:

• one or more network interfaces;
• a processor coupled to the one or more network interfaces and configured to execute one or more processes;
• and
• a memory configured to store a process that is executable by the processor, the process when executed configured to:
• associate an application performance of an online application with network configuration changes implemented across one or more software-defined networks;
• storing, by a device, a plurality of subgraphs, wherein the plurality of subgraphs are extracted from the one or more software-defined networks to comprise network devices and paths linking the network devices of respective different portions of the one or more software-defined networks;
• jointly train using the plurality of subgraphs, based on differences in configuration of different portions of the one or more software-defined networks, a first model comprising a generative network that generates possible configuration changes that includes routing or policy changes and a second model comprising an evaluation network that predicts effects of the possible configuration changes on the application performance for any given portion of the one or more software-defined networks;
• identify, by using the first model, a first subgraph of plurality of subgraphs that is similar to a second subgraph of the plurality of subgraphs, wherein determining that the first subgraph is similar to the second subgraph is based on one or more of:
• a geographic location, a device type, a software version, or a traffic pattern for the online application;
• determine, based on comparing the first subgraph to the second subgraph, difference data;
• generate, based on the difference data between the first subgraph and the second subgraph and by using the first model and the second model, a recommended configuration change for a particular portion of the one or more software-defined networks, wherein the recommended configuration change is generated by the first model and comprises a change to a routing policy, further wherein the recommended configuration change is predicted by the second model to improve performance of the particular portion of the one or more software-defined networks;
• and
• cause the recommended configuration change to be implemented in the particular portion of the one or more software-defined networks.

A tangible, non-transitory, computer-readable medium storing program instructions that cause a device to execute a process comprising:

• associating, by the device, application performance of an online application with network configuration changes implemented across one or more software-defined networks;
• storing, by the device, a plurality of subgraphs, wherein the plurality of subgraphs are extracted from the one or more software-defined networks to comprise network devices and paths linking the network devices of respective different portions of the one or more software-defined networks;
• jointly training using the plurality of subgraphs, by the device and based on differences in configuration of different portions of the one or more software-defined networks, a first model comprising a generative network that generates possible configuration changes that includes routing or policy changes and a second model comprising an evaluation network that predicts effects of the possible configuration changes on the application performance for any given portion of the one or more software-defined networks;
• identifying, by the device and by using the first model, a first subgraph of the plurality of subgraphs that is similar to a second subgraph of the plurality of subgraphs, wherein determining that the first subgraph is similar to the second subgraph is based on one or more of:
• a geographic location, a device type, a software version, or a traffic pattern for the online application;
• determining, based on comparing the first subgraph to the second subgraph, difference data;
• generating, based on the difference data between the first subgraph and the second subgraph and by the device and by using the first model and the second model, a recommended configuration change for a particular portion of the one or more software-defined networks, wherein the recommended configuration change is generated by the first model and comprises a change to a routing policy, further wherein the recommended configuration change is predicted by the second model to improve performance of the particular portion of the one or more software-defined networks;
• and
• causing, by the device, the recommended configuration change to be implemented in the particular portion of the one or more software-defined networks.