The Challenge of Mobile Content Delivery – Part 2
by Jon Bosanac (LinkedIn)
Even with 4G LTE networks, web content transport speeds can be painfully slow compared to CDN-enabled wired networks. nuu:bit has fixed that problem with an innovative technology and service that is purpose-built to address the problem of web content acceleration over cellular networks. But before we describe the solution, we need to explain the problem. This is the second of two blog posts on the causes of slow mobile content transport in the wireless last mile. Today’s topic: RAN congestion.
Network architects often describe radio access network (RAN) congestion as an obstacle for the efficient transport of TCP traffic across wireless networks. This is usually meant to describe the data congestion in the cellular network that impedes the flow of content to and from the end user’s mobile device.
Some of this congestion is due to the unintended consequence by cellular operators in providing network support for the normal operation of the radio medium access control (MAC)-layer which mediates the flow of content over the shared physical radio layer.
This layer not only manages the use of the radio for transmission of content, it also has the ability to retransmit data that has been lost over the air because of low signal strength, interference, inter-cell transfers, or other reasons. In order for this function to work properly, content needs to be buffered so that it can be retransmitted if required.
The MAC layer utilizes retransmission procedures that can successfully and accurately transfer data through the use of radio retransmission procedures without requiring TCP to handle retransmissions at a higher layer.
Because of the highly dynamic nature of radio transmission, data packets are queued in a server associated with the cellular radio transmitter in a periodic manner, waiting for the transmission cycles of the radio MAC layer that controls when the packets are transmitted over the radio medium.
Radio conditions are constantly changing, and mobile devices need time to transition their radios to active status, so data packets may be queued for unpredictable lengths of time. When a mobile user transfers from one cellular zone to another, a radio handoff occurs, which may also add latency. Device queues have been designed to be large enough so that no packets are unintentionally discarded while waiting for the next transmission cycle or radio handoff.
Another significant source of latency in LTE networks is related to what is called “radio resource control (RRC) promotion delay,” which is the time required to transition the user’s radio from a wait state to an active transmit/receive state so that it can participate in data exchange with the cellular base station.
With 3G networks, which are still a predominant cellular network type, this delay is as much as 2 seconds. In LTE networks, the delay is shorter, but can still range from 320 milliseconds to 560 milliseconds.**
The support of the MAC layer operation can be one cause of the large buffering that occurs in the cellular network. Not only can this buffering be large, it can rapidly vary in time, making general assumptions or specific time-based measurements irrelevant.
Because packet loss is either prevented by the large buffers, or packet loss is concealed through the operation of link-layer retransmissions in the cellular network, the TCP sender receives fewer indications of packet loss and will continue to increase its sending rate. This can result in nearly filling most of the buffer space, largely increasing the total RTT, experienced by the TCP flow, and thereby dramatically reducing the efficiency of data flow.
Some mobile operators have installed performance-enhancing proxies (PEPs), for several reasons, including the protection of the cellular network from wide swings of incoming traffic. These PEPs can shield high levels of packet loss and large swings in traffic flow density.
Web servers or CDN edge-servers usually cannot easily detect the presence of a PEP in the data path, and operate in a manner consistent with the assumption that the receiver of the mobile content is the end user device. In order to provide this functionality, PEPs must utilize deep buffers, which when full or near full, can present significant additional latencies to mobile content flow.
Another common characteristic of cellular networks is the asymmetric bandwidth distribution of the uplink versus the downlink directions. Conventional TCP congestion-control algorithms provide throughput that is not only a function of the link and the traffic characteristics in the direction of data transfer but is also impacted by traffic going in the reverse direction. In fact, studies have demonstrated that link and traffic characteristics in the reverse direction have a significant effect on forwarding direction throughput.
This then is a major contributor to inefficient data flow through cellular networks because TCP is not operating to efficiently utilize all available bandwidth. There may be significant bandwidth capacity in the cellular network that is not being utilized but could be with the properly designed transport technology.
By utilizing a transport mechanism that is specifically targeted to optimize flow through cellular networks, these difficulties can be bypassed or mitigated, with consequent acceleration of mobile content through the cellular network and to the mobile user device.
There is a solution to the latency in the mobile last mile – nuu:bit.
**J. Huang, F. Qian, Y. Guo, Y. Zhou, Q. Xu, Z. M. Mao, S. Sen, and O. Spatscheck. An In-depth Study of LTE: Effect of Network Protocol and Application Behavior on Performance. In Proceedings of the ACM Special Interest Group on Data Communication (SIGCOMM), 2013.