All bits are not created equal –

The need for quality of service technology has increased as enterprise networks with different characteristics and priorities such as SNA-based transaction traffic, realtime multimedia, and Web browsing are combined to create a single enterprisewide multiservice utility network.

In the future, voice will also ride the data network infrastructure, providing an even greater need for quality of service that's adequate for realtime voice traffic.

Before we dive into a discussion of quality of service alternatives, let's define our terminology. Originally, quality of service had a distinct meaning associated only with ATM technology. Today, however, QoS commonly applies not only to ATM's cell-based transport but to frame-based environments as well.

Some standards work makes a distinction between QoS and class of service. QoS traffic categories can be fine grained, sometimes specifying specific network performance parameters for individual applications of individual users. In contrast, CoS organizes services into classes that are treated separately, often only by prioritization. (That kind of CoS is not the same as IBM's SNA networks' class of service standards.)

Strictly speaking, CoS is a subset of QoS. CoS is simpler to implement, particularly on multihop networks, and it scales more easily to large networks. Recent IETF QoS work has concentrated on Differentiated Class of Services to provide a scalable method for grouping traffic flows for different levels of service quality.

In common usage, QoS encompasses CoS.

To develop a comprehensive technical approach to QoS, you must evaluate and select alternatives differently for LANs, which usually offer abundant free bandwidth, and WANs, where bandwidth is more constrained or more costly.

In the WAN, there's usually more traffic than bandwidth available, so considerable contention can occur. There is no "one size fits all" QoS recommendation for supporting all applications or traffic types over a WAN. The choice of appropriate QoS options depends upon the applications or traffic types on the network.

Network traffic can be coarsely grouped into three categories, differentiated by how much latency (network delay) they can tolerate:

Realtime traffic such as conversational voice, videoconferencing, and realtime multimedia requires very short latency (typically less than two-tenths of a second end to end, including processing at the end stations) and controlled latency variation (jitter).

Transaction traffic such as interactive transaction processing, remote data entry, and some legacy protocols, including SNA, require latencies of approximately one second or less.

Bulk transfer traffic accepts virtually any network latency, including latencies on the order of a few seconds.

Within each traffic category, we can further subdivide traffic by priority. Top priority traffic receives preferential treatment because of its importance to the enterprise. Priorities are not substitutes for traffic categories, which are absolute, not relative, requirements. As there are no absolute network measurements associated with a priority, the meaning varies from one network to the next in terms of measurable network performance of a particular priority. Priorities can be used to differentiate among user groups or to differentiate among applications and users within a group.

Your service provider or the WAN manager in a private WAN adjusts the amount of capacity available within each traffic category to meet an agreed-upon service-level agreement for each user group. You need to specify priorities in the routers only if you lack sufficient capacity within a category to accommodate all the traffic.

Next time we'll talk about your alternatives for providing QoS across the WAN.

This story, "All bits are not created equal" was originally published by ITworld.


Copyright © 2000 IDG Communications, Inc.

7 inconvenient truths about the hybrid work trend
Shop Tech Products at Amazon