Experiments on Network Services for Video Transmission using FABRIC Instrument Resources.

Video-based applications form one of the most popular applications on the Internet that is continually evolving. There is a need to develop novel network services that enable reliable video transmission over network paths with dynamic cross-traffic, as well as services that utilize programmable data planes enabled by Protocol-independent Packet Processors (P4). In this paper, we describe experiences in developing network services for reliable video transmission using resources from the FABRIC network instrument, which supports high-performance edge/cloud as well as programmable networking infrastructure. Specifically, we deploy a programmable network using local Ethernet (Layer 2) sites, as well as geographically distributed wide-area network (LAN extension) sites in FABRIC, in order to experiment with visual cloud computing application use cases. Our experiment results provide insights into benefits of data plane programmability (i.e., port forwarding) on improving video streaming quality and compare CPU vs. GPU processing times while completing object detection pipeline processing to obtain visual situational awareness.