Performance Within A Fiber-To-The-Home Network

By: 
M. Sargent, M. Allman
Appears in: 
CCR July 2014

Fiber-To-The-Home (FTTH) networks are on the brink of bringing significantly higher capacity to residential users compared to today'ss commercial residential options. There are several burgeoning FTTH networks that provide capacities of up to 1 Gbps. We have been monitoring one such operational network the Case Connection Zone - for 23 months. In this paper we seek to understand the extent to which the users in this network are in fact making use of the provided bi-directional 1 Gbps capacity. We find that even when given virtually unlimited capacity the majority of the time users do not retrieve information from the Internet in excess of commercially available data rates and transmit at only modestly higher rates than commodity networks support. Further, we find that end host issues - most prominently buffering at both end points - are often the cause of the lower-than-expected performance.

Public Review By: 
Fabián E. Bustamante

Fiber-To-The-Home (FTTH) networks have started to bring potentially game-changing increases on residential network capacities. From Stanford and Kansas City to Austin and Cleveland, a number of services are offering upward of 1Gbps to some lucky residential users -- whether they and their applications can leverage it, or not. This work is an early traffic measurement study of a production FTTH to answer a version of this last point – can protocols and application implementations take advantage of the provided capacity? The authors tackle the question, focusing on TCP performance, using packet-level traces collected, over two years, at a small (90 users) residential deployment known as the Case Connection Zone. While limited in scale and focus, this is the first study of capacity utilization in an operational FTTH network. Perhaps not surprisingly, the study finds that users’ demand on the network remains within commercially available data rates, leaving much capacity available for yet-to-imagine applications and services. Their analysis also suggests that TCP implementations may be artificially limiting performance as, at least theoretically, hosts should be attaining much higher throughput than what they have observed. A first round of reviews challenged the authors to go deeper in some of their analysis and be more explicit in the scope of their argument and contributions. The final version is a nice contribution that shedslight on some of the real issues that need to be addressed to make effective use of FTTH capacities. All reviews were very positive about this as a first attempt at the problem, clearly opening an interesting research agenda. Going deeper into the remaining undiagnosed bottlenecks and considering alternative protocols are two of the many open lines of inquiry.