Bandwidth management pays off: quality-of-service measures guarantee application performance - Service providers/service: quality of service

Communications News, Nov, 2001

Today's enterprise often consists of a number of remote offices and branches, interconnected with a hodgepodge of access technologies and bandwidth speeds. This translates into more employees having to access remote servers across wide area networks (WANs) and extranets as part of their daily routine.

In the absence of a viable scheme as to how users can reliably and consistently access these applications, more of them will face increasingly long application-response times and, worse, will face timed-out connections. To prevent this, network managers are turning to bandwidth/quality-of-service (QoS) management as one of their tools for successful application delivery.

Guaranteeing QoS for a particular application becomes problematic at network congestion points, particularly the local area network (LAN)/WAN interface. Application traffic in a bandwidth-rich LAN environment must be aggregated into the smaller WAN links and then carried to multiple destinations. As a typical example, Internet traffic passing from the branch office through the headquarters and then to the ISP has an ability to limit bandwidth available for critical applications (e.g., service access point) that may simultaneously need access to headquarters servers.

To alleviate the problem, network managers have two choices--add bandwidth or provide bandwidth management. Adding bandwidth to a WAN link is expensive and is, at best, a temporary fix. Because no additional control is added, further bandwidth upgrades will be required when utilization eventually increases. Adding bandwidth management is actually the most cost-effective technique to ensure the required application behavior. Depending on how much control is needed, bandwidth management can range from using QoS mechanisms in existing devices or adding dedicated bandwidth-management devices for better control of the network. Voice over Internet protocol (VoIP), for example, has specific bandwidth, jitter and latency requirements, and dedicated bandwidth-management devices become the required solution.

KNOW YOUR NETWORK

The most effective tool in ensuring proper application performance across an enterprise network is to monitor how the network is being used. Network usage monitoring can help pinpoint network congestion points, spot users abusing bandwidth and identify how applications are using bandwidth. In addition, how networks are used is constantly changing, and knowing how traffic patterns are changing can form the foundation for making sound network configuration changes.

For example, uplinks are typically considered candidates for upgrade when they consistently reach 70% to 80% utilization, but if network usage reports show that Napster is using 40% of the bandwidth, applying bandwidth-usage policies will be more cost-effective than upgrading equipment and bandwidth. As new real-time communication applications are being introduced, such as VoIP or videoconferencing, new requirements are being added where bandwidth has to be simultaneously guaranteed for several applications.

Once needs have been assessed, devices must accordingly enforce bandwidth-management policies. Each of these devices, however, differs from vendor to vendor, and the protocols that allow some level of interoperation are just emerging.

The techniques that are used have different trade-offs and have real effects on network performance. The more sophisticated techniques give more definite control over bandwidth usage, but devices, particularly routers that already have complex tasks to execute, may encounter performance issues due to increased packet processing overhead. In addition, most bandwidth-management techniques, such as rate limiting (also known has policing), result in dropping packets, which negates any advantage in bandwidth-management control.

In general, standalone bandwidth/ QoS-management devices should be used for than the most basic WAN output queue QoS needs, for example, in converged voice, video and data applications where simple output queuing mechanisms cannot simultaneously guarantee bandwidth availability. Careful consideration should be given to how these devices enforce QoS, and careful testing in a real-world network should be examined.

Devices enforce bandwidth policies in two ways: queuing techniques and traffic shaping/congestion avoidance mechanisms. Queuing techniques dictate in what order packets are delivered to the WAN interface and can help deliver priority applications through the network. They range from simple first-in/first-out queuing, with no bandwidth control, to class-based queuing with precise bandwidth controls. Class-based queuing should be the preferred technique, as it is the only one able to provide the advanced prioritization required for a deterministic approach of bandwidth guarantees.

MATCH ENFORCEMENT WITH NEEDS

Traffic shaping and congestion avoidance are ways to prevent network congestion by using several techniques that signal transmission control protocol--the most common transport protocol in enterprise IP networks--to slow and control communicating flows between devices. Most implementations work by monitoring the priority queues and applying one technique when the queues reach a set threshold.