World Wide Packets 

Service Assurance Hard QoS

Introduction

Quality of Service (QoS) is one of the most widely abused terms in the telecommunications industry today. Many vendors claim to achieve it, but the actual implementation and capabilities vary considerably. World Wide Packets’ solution includes a set of simple, but very powerful, mechanisms that enable the industry’s best QoS for multi-service Ethernet access networks. To date, such effective QoS techniques have been offered only within expensive core network platforms. World Wide Packets is the first company to enable this Carrier Class level of QoS in a cost-effective and scalable access environment.

Current QoS Implementations and Limitations

Congestion in a network exists when, at a given point in time and at a given location of the network, the ingress traffic exceeds the egress capacity of the network. This typically applies to an aggregation network where several downlinks are aggregated together into one or more uplinks. A number of basic mechanisms have been implemented to address congestion.

One mechanism is to queue the ingress traffic until the egress congestion is resolved. Of course, such a method is limited by the amount of queuing memory the network device offers. Increasing the queuing capacity directly increases the cost and the complexity of the device. Although queuing frames may prevent loss, it also increases latency and latency variation (jitter) to the traffic through the network. This latency may not significantly impact data traffic, but can have very negative impacts on real-time traffic, such as voice traffic.

Another mechanism is to drop excess traffic in a pseudo-random fashion as it arrives at the congested network device. Higher layer protocols such as TCP can sense traffic loss and request retransmission for some data applications, but other applications such as video, are highly sensitive to any information loss, and can be rendered inoperable by this mechanism.

A variation and improvement on the pseudo-random mechanism permits distinguishing critical traffic from non-critical traffic, then discarding non-critical traffic first during congestion. However, at a certain point of congestion, even critical traffic will be discarded. This method works best when the majority of the traffic supported by the network has been defined as non-critical.

All of these QoS implementations evolved over time in order to better support critical data applications, but have proven to be ineffective when the network must support loss-sensitive applications such as video. Today, network operators need to be able to reliably deliver any service, to any subscriber, with the QoS that is necessary for each application.

The following paragraphs provide an overview of some of the more common methods of achieving QoS today.

Rate Limits

Rate limiting, one of the most common methods used to implement QoS today, is also the most inadequate. With rate limiting, a network operator provisions the maximum amount of bandwidth that each subscriber is allocated. By ensuring that the sum of all the rate limits is not more than the total available bandwidth, a network operator can “guarantee” that every subscriber will be able to access their bandwidth at any time.

With rate limiting the network is provisioned for the worst case; i.e., every subscriber peaking at the maximum bandwidth allowed all at the exact same time. Thus, most of the time, when subscribers do not peak at their maximum bandwidth or do not peak at the same time, the network is grossly underutilized. Huge amounts of bandwidth that could have been assigned to new subscribers are wasted in order to handle the worst-case scenario. This leads to steep revenue loss, delaying the return on investment or even prevents attaining break even on the investment.

Link Utilization equals Revenue Generation. In the diagram below on Basic Rate Limiting, each link is capable of supporting 1,000 Mb/s. The maximum traffic on link A-C is 600 Mb/s, or 60% of the total up link capacity and results in stranding 40% of the revenue potential of the link. The maximum traffic on link B-C is 400 Mb/s, or 40% of the total link capacity and results in stranding 60% of the revenue potential of the link. Rate limiting results in significant amounts of underutilized bandwidth that severely limits the revenue potential of the network.

Figure 1. Basic Rate Limiting

Class of Service

The next commonly employed method offers little improvement to the first. IEEE defined priority in its 802.1p standard (now incorporated in the latest 802.1D standard) to help reduce the delay introduced by forwarding and queuing within a network. While this is a very effective mechanism to reduce latency, it becomes highly inefficient when applied to “guaranteed” bandwidth.

The standard uses priority to decide which traffic to discard first in case of congestion and actually offers a hierarchical way of dropping “guaranteed” bandwidth. Dropping guaranteed bandwidth is obviously the last action a network device should take when congested and should not be an advertised feature.

In the diagram below, device C ends up dropping 500 Mb/s of priority 0 (P0) “guaranteed” bandwidth. Note that although the priority offers up to 8 different values (0-7), the priority 7 is usually reserved for provider management traffic.

In addition, a priority mechanism applied to guaranteed bandwidth only works when end-users actually subscribe to different levels of priority. If several subscribers choose the same priority, whatever the priority value is, it will be impossible for the device to drop “guaranteed” bandwidth in a hierarchical order, resulting in random drops of traffic.

Figure 2. Basic Rate Limiting with CoS

Continue