dc.description.abstract | In recent years, the concept of edge computing comes up. Due to the access of many IoT devices to wide-area networks, although the computing power of single node is weak, it offers great possibility when they are grouped. However, as more and more devices and new network protocols appear in the WAN, traffic classification becomes more and more difficult. Many researches attempt to use deep packet inspection, machine learning or domain name system (DNS), which are insufficient for encrypted packages and cannot be directly deployed in real network environments.
The FIPAC packet classification mechanism proposed in this paper are intentionally deployed on edge networks for the reason that more fully qualified domain names (FQDNs) are obtained more easily in comparison to core networks. Then, the FQDN can be classified with effective labels fetched from Wikipedia entries and autonomous system number. This is how FIPAC achieves automatic classification and fast inference. Compared with other classifiers, the FIPAC mechanism is more user-friendly and efficient, also, requires less computing resources and respects users’ privacy. With FIPAC, network operators do not need to worry about exhausted computing resources taken by packet classifiers. They can focus on the utilization of classification results, such as QoS control on distinct applications.
In this paper, we deploy FIPAC on software router on real edge networks. Compared with commercially available embedded edge routers and other types of classifiers, we verify that FIPAC takes advantages in performance and flexibility. | en_US |