Real-time Virtual-network-trafficmonitoring System with FPGA Accelerator

There is a growing demand for a virtual-network-traffic-monitoring system for managing and controlling network services. Such a system must be able to handle high-load processing such as analyzing encapsulated packets and network-traffic classification using many header fields. To visualize network traffic for a virtual machine in real time, we propose a real-time virtual-network-trafficmonitoring system with a field-programmable gate array (FPGA) accelerator. Our system consists of a resource-saving hash-based network-traffic classifier (NTC) that classifies virtual network traffic at high speed using many search conditions. The hash-based NTC reduces memory resources by using a twostep hash search. Our system with this hash-based NTC provides a real-time visualization of multiple statistics such as the number of packets, bytes, microbursts, and histograms of jitter and latency for each virtual machine. To verify the performance of the hash-based NTC, we evaluated the number of searches per input packet. As a result of classifying virtual extensible local area network (VXLAN) packets into 10,000 categories using 17 header fields, the average number of searches executed with the hash-based NTC was about one-fourth that of a search-tree-based NTC. In addition, memory and logic-resource usage of the hash-based NTC were on average about 40 and 80%, respectively, which were less than those of several FPGA-based ternary content addressable memories with the same rules. Finally, we demonstrated that our system with the hash-based NTC visualizes VXLAN traffic for each virtual machine in real time. © 2021 Nippon Telegraph and Telephone Corp.. All rights reserved.