Abstract
Monitoring and troubleshooting networks to improve their performance and reliability is a complicated task, not only because it requires the checking of every single network device but also because it involves understanding the connections between those devices. To address that complexity, techniques are presented herein that support a Stratified Investigation of Low-Performing Network Architecture (SINA) system. Such a system is a framework that identifies any low-performing network architecture areas and makes improvements on a subnetwork level. Such a system may automatically identify low-performing areas of a customer’s network based on information about similar networks and expert knowledge with solutions. Such a system may employ a graph neural network (GNN) to identify areas of a customer’s network that need improvement based on a calculated performance score while considering the interaction between devices and the topology of networks. Further, such a system may leverage network performance metrics from many customers to create a performance benchmark and then evaluate where a customer’s network’s performance lies within that benchmark. Still further, such a system may employ a transformer-based natural language processing (NLP) model (that understands key semantic knowledge from documents, device configurations, and logs) to help generate solutions to network issues. Finally, based on high-performing customers relative to the benchmark and documents with best practices for configuring networks, a SINA system may provide solutions to a customer’s network that will help optimize network performance.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Chang, Benjamin L.; Gong, Qixu; and Shao, Qihong, "STRATIFIED INVESTIGATION OF LOW-PERFORMING NETWORK ARCHITECTURE (SINA) USING GRAPH NEURAL NETWORKS AND PEER BENCHMARKING", Technical Disclosure Commons, (March 22, 2023)
https://www.tdcommons.org/dpubs_series/5749