Current Filter: >>>>>>

   

In the age of leased circuits, application quality of service across the WAN was straightforward. At the point of WAN access, first mile traffic shaping methods delayed low priority bulk traffic types, offering high priority, critical traffic faster access. The last mile and the first mile link were tightly bound and had an implied backflow which prevented overrunning the last mile. At times when there was no higher priority traffic, traffic shaping methods would allow lower priority traffic access to all available bandwidth. Predicting future WAN utilisation was straightforward as applications were relatively closed to non-supported enterprise applications, traffic flowed through fixed hierarchies and was mostly transaction oriented. Purchase decisions for bandwidth and policy provisioning needed to be adjusted relatively infrequently.

The arrival of oversubscribed packet switch networks like MPLS and mesh topologies ended this simplicity. For example, two first mile sites could potentially overrun a last mile site's available bandwidth, resulting in an unpredictable and potentially undesirable end user network quality experience.

This problem has been exacerbated by the growth in less predictable and more demanding applications. Real-time applications like VoIP and Unified Communications platforms can add unexpected traffic spikes. Increased Internet usage in offices, including the BYOD trend, has added to bandwidth demand and the level of unpredictability.

Service providers generally provision service levels that, in theory, help mitigate these issues with differing classes of service that are optimised for differing traffic requirements. For example, Voice over IP applications typically utilise an Express Forwarding (EF) class of service to minimise congestion and jitter by strictly policing traffic rates. This policing typically occurs at the last mile. In a network where several sites are sending EF traffic to a single site, the lack of coordination among the senders can overrun the provisioned EF rate at the receiving site and cause poor performance of the Voice over IP application.

This type of fixed service provisioning leads to wasted bandwidth at the first mile, as traffic may be transmitted over the WAN that will be discarded at the last mile because of oversubscription and congestion. With the increase of VoIP, VDI and mesh collaboration applications that are sensitive to latency, jitter and packet loss, it is more difficult to model and predict which service levels are required at the first and last miles. The complexity is even greater if the service is provided across multiple geographies, especially internationally.

The variability in service offerings also increases complexity. Different providers implement QoS differently and these differences can affect application behaviour, especially during times of high congestion. If a company has to use more than one service provider or has locations where MPLS isn't available and thus broadband links are used, the prioritisation behaviour will vary from one location to another. Understanding the method used by each provider in order to correctly size and provision the network becomes a daunting task.

In other words, the network and the applications are just too dynamic and complex for the traditional fixed service provisioning approach. One approach to this complexity is to just overbuy bandwidth and prevent the need for service level policies. This approach is not viable for most enterprises because of the cost. This approach also fails to provide any prioritisation of services in cases where network failures cause a reduction of the available bandwidth.

The addition of inexpensive, best effort broadband as an additional source of bandwidth leads to more complexity as the quality of the bandwidth is less predictable. What is required is a more intelligent and adaptive approach - one that utilises bandwidth reservation, first mile to last mile synchronisation and active adaptive service level adjustments, all combined with real time application and network monitoring.