UC Santa Cruz

A fundamental goal of mobile network service providers is to guarantee the networking services' reliability, which includes two major goals: reliable network access and security of the communication channels. As today's mobile network is getting larger and more complex, traditional protocols and systems for satisfying the above goals, which are mainly based on rules and experiences, are no longer scalable or effective. On the other hand, together with the growth of the mobile network scale is the growth of the mobile service data. The rich context and knowledge behind the large-scale network service data can provide us with new opportunities to understand the network states and support the design of reliable network systems. Therefore, this thesis research mainly focuses on exploring and designing data-driven approaches to enhance mobile network service reachability and security. Specifically, this paper focuses on two challenging problems faced by the mobile service providers regarding network service reliability.

The first problem is how to automatically and efficiently recognize the root cause of a network accessibility problem experienced by the end-device users in cellular network services. Specifically, we design and implement an automatic troubleshooting system named NeTExp. NeTExp learns to identify the root causes of a user-side service problem through deep neural networks (DNNs) that are capable of extracting the complex spatial-temporal features of the massive cellular network log data. It also uses advanced weakly-supervised learning methods for training the models and thus overcomes the data limitation challenges in practice. The system is trained and validated using an extensive period of network and customer care data from a major US cellular service provider.

The second problem is how to enable low-cost and fast peer-to-peer authentication among a large mobile network ecosystem, such that secure mobile channels can easily be guaranteed for any pair of devices in the network. The critical challenge is that many mobile devices, especially IoT devices, do not have the capability to maintain the certificate revocation status for a vast device universe during the authentication process. To solve this problem, we design a fast on-device authentication prototype called TinyCR. TinyCR utilizes super-efficient data structures to maintain the certificate revocation status on each device locally. It also enables fast synchronization in response to any changes in the certificate universe. Through evaluation, we show TinyCR outperforms other state-of-the-art certificate revocation checking protocols regarding memory cost, checking efficiency, and synchronization cost. Those new features of TinyCR can well enhance the peer-to-peer channel security in a large mobile network.