Evaluation of Background Transport Protocols in Production and Experimental LTE Networks.

As cellular networks are becoming faster, cheaper, and wider in terms of reach, it is today the carrier of ever-increasing bulk of network traffic. Cisco has predicted mobile data traffic to exceed 77 exabytes per month by 2022. Based on the time critical nature of traffic, it can be classified into high priority foreground (multimedia, real-time app data) and lower priority background traffic (software, firmware updates and cloud-sync) and it is highly undesirable for foreground flows to be affected by competing background flows in the same bottleneck link. Network policy dictates that such lower priority traffic should be transmitted only using spare capacity. State of the art congestion control schemes for background traffic achieves this by using one-way delay or round trip time as an indicator for congestion. This is an effective solution in wired or Wi-Fi networks, however they are inefficient in cellular networks due to operational characteristics of cellular scheduler (Proportional Fair (PF)), that aims at providing fairness at short time granularities. We present Legilimens, an end to end transport for low priority background flows in cellular networks. Legilimens leverages the traits of PF scheduler to detect contention in the last hop. This thesis evaluates Legilimens in real LTE production network and in emulated LTE test setup: Phantomnet. We compare Legilimens with several background transport protocols like LEDBAT, TCP-LP, and Vegas and with popular congestion control schemes like Cubic and Reno using specifically designed test methodologies like, one-on-one and mixed workloads. We demonstrate that Legilimens efficiently yields to foreground traffic and is agile enough to capture spare bandwidth when available.