Flow-based Service Type Identification using Deep Learning

Abstract

Automatic identification of the service type used by network flows (e.g., HTTP and MySQL) is an essential part of many cloud management and monitoring tasks for quality of service, security monitoring, resource allocation, etc. Several studies have adapted deep learning models for accurate service type identification of network traffic. These models vary in how the message flow data is used and what datasets are considered. There are no published guidelines on selecting the best approach for automating the service identification process. In this paper, we opt to fill such a technical gap and provide a detailed study of the trade-offs of different deep-learning based approaches for service type identification of network traffic. Towards this end, we generate flow-based datasets for a wide range of service types that are commonly deployed in the cloud. We consider two different deep learning models that have shown promising results in this context, and show their performance for both payloadand header-based datasets, considering fundamental parameters such as dynamic service port configuration, flow direction and the packet order in the flow stream.