Quality of service has emerged as a salient architectural element missing from today's IP-based backbone networks. Yet it's clear that, so far, the lack of QoS has not dimmed IP's prospects for global domination. So how seriously should you take warnings that, without a workable standard for QoS very soon, prospects for converging IP-based data and voice traffic are doomed?
Chances are that right now you don't need QoS on your LAN. If you're like most of your contemporaries, you're still operating mutually exclusive voice and data networks. The bulk of your switched voice traffic is sent to an interexchange carrier, to which you pay something like four to seven cents per minute per call for delivery nationwide.
But you may have also noticed that, in any given week, a few more of the Fortune 500 are announcing big-time migration plans towards convergence. Do they know something that you don't?
Not necessarily. Most of these enterprises are planning to achieve convergence via voice over IP, and VoIP does not require any special QoS as long as the IP backbone is well behaved -- that is, as long as there is no congestion anywhere along the transport route that would lead to buffering.
VoIP uses an efficient mechanism called a jitter buffer to adjust each realtime VoIP stream to accommodate slight temporal variations in transmission. Packets may skew in time by 30 or 50 or even 100 milliseconds due to the dynamic nature of IP packet transport. The jitter buffer realigns and resynchronizes the packets.
However, if you hit a router interface that is congested -- such as a 100 Mbps Fast Ethernet LAN funneling traffic onto a T1 WAN link -- the result will be buffering (or dropped packets). Buffering will add at least 100 milliseconds of one-way delay to VoIP packets, and consequently will lead to those packets being discarded at the receiving end.
If you want IP convergence without standard QoS, your choices are either to avoid congestion and buffering -- easier for enterprise networks than for carriers and service providers -- or accept that, from time to time, there may be congestion and buffering, and VoIP quality will drop whenever that happens.
But those aren't your only choices. You can also implement one of the current nonstandard forms of QoS on your IP backbone. At Mier Communications, we've tested various IP prioritization techniques and products. For example, Cisco's latest IOS offers new features that implement prioritization policies -- sort of -- in the form of access control lists. And numerous standalone-device vendors, like Packeteer, offer boxes that effectively allocate bandwidth to match user-specified traffic classes, as well as rearrange packet streams on the fly so that high-priority packets always go ahead of others.
You can deploy whatever QoS technique and product you want for your backbone, whether you are an ISP or an enterprise. But interoperability among the various products is still far from assured. That becomes more of an issue when the same traffic traverses both LAN and WAN. However, there are already QoS boxes that operate more or less transparently, so interoperability need not be a great concern.
The IETF continues to advance overlapping and competing draft standards -- Resource Reservation Protocol, MPLS, DiffServ, and several others. We see DiffServ as offering the most viable, practical, and deployable solution. But even that is still two years away from what would be considered widespread deployment.
For now, while you're monitoring developments on the standards front, teach yourself about realtime traffic types like VoIP with regards to bandwidth and packet structure. If you must, you can implement any of several QoS products that are now on the market. Just be sure to find the one whose capacity, scope of QoS control, and alignment with the emerging QoS standards best suit your network.