Why you need a Qos scheme
W ithin weeks of each other, 3Com and Cisco have announced their plans for delivering policy-based management - dubbed, respectively, the 3Com Transcend Policy Manager and CiscoAssure Policy Networking. Both companies have targeted class of service (CoS) or quality of service (QoS) as the first service they'll bring under their policy-based management umbrellas.
It's great that network managers will be able to centrally define policies with regard to CoS/QoS. However, it begs the question: Who needs CoS/QoS, anyway?
There's a long-held belief that in campus networks, bandwidth can solve all problems. Gigabit Ethernet and 622M bit/sec ATM should provide enough bandwidth for anybody, right?
Maybe or maybe not. Think back to fall 1996. Are you running the same applications you were then? How much has your traffic grown since then? How are your response rates?
If your network handles data only, traffic has grown slowly and users rarely complain of poor response times, you're probably in good shape. Adding bandwidth where needed will likely keep your network humming until you have a major change in your application mix or in the number of users on the network.
However, if your traffic has grown by 35% or more or response times have increased by 10% or more and you see these trends continuing, you should seriously consider some type of CoS/QoS scheme. The alternatives are to keep throwing bandwidth at your problems or freeze your applications and associated traffic at their current levels.
Certainly more bandwidth will make data move faster. However, it won't magically make congestion go away or prevent router and switch buffers from overflowing. When one or more packets have to be dropped, do you care if packets from the order entry application get dropped rather than e-mail packets? If you consider the mismatch between gigabit-speed backbones, 100M bit/sec links in building risers and 10M bit/sec links to desktops, you're bound to have congestion points.
More important, adding bandwidth does nothing to change the fact that LANs provide only a best-effort, one-size-fits-all service. Consequently, your SAP R/3 traffic can drown in a flood of PointCast packets, and packets from a video training session can get blocked by a Web download. If you want better control over which traffic gets through, you need some sort of CoS/QoS scheme.
If yours is one of the small but growing number of organizations that are rolling out interactive applications, such as videoconferencing or collaborative white-boards, you will likely need some type of CoS/QoS to ensure smooth voice and video delivery. You could get by without CoS/QoS if your network consistently is lightly loaded. However, the bursty data traffic typical of LANs can easily clobber real-time applications such as videoconferencing, making users unhappy.
Fortunately, the industry has developed a few schemes that make CoS/QoS support in the LAN relatively painless. The simplest of these is a prioritization scheme. The IEEE and Internet Engineering Task Force support the same eight levels of priority - the IEEE in its 802.1p specification and the IETF using IP precedence bits in the IP header.
3Com and Cisco are among the many vendors moving quickly to support these prioritization schemes in their switches and routers. In addition, the companies are among the vendors that have worked with Microsoft to ensure that the next releases of its operating systems have the hooks needed to set these priority bits.
Coupled with sufficient bandwidth, these priority schemes will give you the control you need to expedite certain types of traffic. For example, with a prioritization scheme in place, you can define SAP R/3 traffic as high priority so it will be forwarded before PointCast and other low-priority traffic. And if packets must be dropped because of congestion, the low-priority packets will be dropped first.
For organizations that need to control latency, there are more elaborate QoS schemes, such as those supported by ATM and the Resource Reservation Protocol (RSVP). These QoS schemes give you control of bandwidth, latency and accuracy levels (meaning which packets get tossed in case of congestion). RSVP is capable of ensuring that latency doesn't exceed a specified maximum, while ATM goes a step further and can control jitter - the variation in latency, or delay, that a packet experiences as it moves from one device to another across links.
If you're not sure you need CoS/QoS today but think you might in the future, hedge your bets. Buy switches for your wiring closet and the network edge that have two levels of queuing so you can support 802.1p-based prioritization. Similarly, buy core switches, Layer 3 switches, switching routers and routers with multiple levels of queuing (four has become relatively common) to ensure that you can accommodate more levels of prioritization.
The bottom line: If the performance of your mission-critical applications is suffering as network traffic increases, you should plan your CoS/QoS strategy now.
