KENET: A Bandwidth Management Case Study

By: Kevin Chege, Mat Ford

Date: February 7, 2009

line break image

As part of ongoing efforts to better understand and respond to the variety of networking issues that can loosely be classed as stemming from bandwidth-intensive activities (see “The Internet and Bandwidth-Intensive Activities,” IETF Journal, Volume 4 Issue 2), the Internet Society invited Kevin Chege of the Kenya Education Network (KENET) to attend IETF 73 in Minneapolis. Kevin has considerable personal experience dealing with the impact that bandwidth-intensive applications can have on a relatively low-bandwidth region of the global network. Attending the IETF meeting allowed Kevin to share his perspective with the engineers working in the newly formed ALTO (Application-Layer Traffic Optimization) and LEDBAT (Low Extra Delay Background Transport) working groups (WGs) and to learn from them how these activities might help him in his work.

Established in 1999, KENET connects educational institutions and research centres in Kenya with the goal of distributing knowledge throughout the country and making sure that the research and education communities have access to the Internet. Currently, there are 8 member institutions directly connected to the main node, which is in the capital city of Nairobi, and more than 40 additional member institutions that participate in the network by way of copper leased lines that are made available by Telkom Kenya, one of Kenya’s main backbone providers. The main node consists of a 2 megabits-per-second uplink via a leased line, a 1 Mbps uplink via VSAT, and a 1.5 Mbps downlink via VSAT. The members’ bandwidth usage typically ranges from 64 kilobits per second to 960 Kbps. Their leased lines terminate at KENET on E1 lines with a maximum 2 Mbps capacity. In some cases, members have their own VSAT downlinks due to the limitations of their leased line and use KENET only for uplink capacity.

To address some of the bandwidth limitations, KENET has begun working on a World Bank-sponsored project aimed at improving access for its member institutions by migrating from a copper-based infrastructure to fibre. This project is expected to reach completion by late 2009. Although the initial bandwidth purchase will be on VSAT, it is hoped that KENET will benefit from the price reduction and increased bandwidth availability that may be possible when undersea cables arrive in 2009. However, even when the cables become operational in mid-2009, it is expected that Internet bandwidth will be in short supply due to both financial pressures and limitations to the physical infrastructure.

The sort of problems and issues Kevin sees in his day-to-day work include WAN links from member institutions being saturated with P2P traffic, sometimes to the almost total exclusion of other traffic. This naturally stifles use of the Web for productive research by students and staff at the institutions served by KENET. Other important issues are related to (1) the lack of well-formulated IT policies that could help manage specific types of applications on the network, (2) the need for more detailed and widespread network monitoring, and (3) the need for better training of local network administrators, many of whom know what they want to achieve but are less clear on how to go about achieving it. This knowledge imbalance often leads to ineffective and, in some cases, counterproductive solutions being deployed.

One of the more successful strategies that has been adopted over time by KENET is regular training for staff on subjects like network management, security, and network monitoring using opensource tools. More-aggressive use of access-control lists and the widespread deployment of Web caches, spam, and antivirus filters have also helped mitigate many of the original issues. Detailed proposals have been developed to integrate a bandwidth-management and optimization tool for deployment at all member sites. This solution, which is yet to be developed, aims to provide a simplified interface to free, opensource tools, which will enable local network administrators to monitor and manage their local connectivity without needing highly specialised training beforehand. Having a standard solution for member network bandwidth management would also ease KENET’s job of supporting the member networks remotely. It is hoped that such a solution could be applied to other upcoming National Research and Education Networks in the region, many of which could experience similar problems as they develop.

KENET and the member institutions want the flexibility to be able to push back on certain bandwidth-intensive applications at certain times of day, for example, when the network is under simultaneously high demand from researchers and staff trying to do their research at the educational institutions. From this perspective, the sort of solution that the LEDBAT WG is working on should benefit KENET. The LEDBAT WG charter identifies a common scenario wherein applications experience large delays in the presence of P2P applications uploading over thin home uplinks; for KENET its entire WAN is peppered with thin uplinks. The potential benefits of a deployable solution in this space are therefore much greater than just the well-known use-case of an ADSL user simultaneously trying to run a P2P application and place a VoIP call. (It is less clear that the direction being pursued by the ALTO WG will be as useful for the KENET situation given that KENET’s goal isn’t so much keeping P2P traffic on-Net as it is ensuring P2P traffic gets out of the way whenever there are other, more-interactive applications trying to share the network resources. Because most content desired for download by P2P clients in Kenya is presumably off-Net, the sort of policy tools that an ALTO solution could provide may be ineffective in this instance.)

Speaking to this point directly at IETF 73, Kevin said he had observed a lot of engineers at the IETF making assumptions about the capacity of enduser connections based on their own, narrow domestic experiences. In reality, users in low-bandwidth networks have habits that are similar to those in high-bandwidth networks. They want to visit social networking sites, they want to download software and multimedia files, and they want to play online games, all of which may be bandwidth intensive and could effectively cripple the network for other users. The bandwidth-to-host ratio, poor-quality bandwidth and infrastructure, and, in most cases, the increased latency caused by the use of VSAT all make this a serious problem for network engineers in Kevin’s home region. Bandwidth-management-related WGs at the IETF should ensure that their solutions consider and address the kinds of problems faced by KENET in its day-to-day network management role.