Carl Weinschenk spoke with Brough Turner, senior vice president, CTO and co-founder of NMS Communications, a provider of mobile technology. Turner recently posted an item on QoS at his blog, Communications.
Weinschenk: Your position is that QoS only is necessary in very narrow circumstances. But we've been hearing about it for years in relation to provisioning advanced and convergent applications. Is there a disconnect?
Turner: I think the reason that everybody is hyped about it is that 10 years ago best effort was fine for e-mail and Web browsing on the Internet. But [in the new era of VoIP] what happens if a giant file transfer occurs when you are trying to talk? There are latency issues if the delay is more than 200 milliseconds. If, for example, there is more than one satellite hop, then you can hear it. People start talking over the person on other end. Beyond 200 to 250 milliseconds, people complain about the quality of the call. If it reaches 500 milliseconds, it sounds like a walkie-talkie. That started all the interest. There was a lot of development and standardization on QoS - but not a lot of deployment, certainly not in the Internet backbone. The one place where it makes sense is where there is limited bandwidth and no economical way of expanding the bandwidth. In the Internet backbone, it is never a problem, which is what I discuss in the blog. In corporate LANs, it is always cheaper to put in more capacity than QoS because QoS is complicated to install and especially to maintain.
Weinschenk: Some people say there are problems - such as jitter and latency - that aren't solved by more bandwidth. Is this so?
Turner: That's only true in very narrow windows. With a lot of capacity, it's possible to solve basically most of the problems. QoS does not add any more bandwidth. It just allows one app to run and stall out the other. There is a very good article by Dan Bricklin, one of the two guys associated with VisiCalc, that I point to in my blog. It explains in non-technical terms why it is cheaper to throw bandwidth at things other than QoS. QoS is good in a narrow window if you can't buy more bandwidth, which is not the case in the core or corporate LANs. If a corporation with a bunch of satellite offices needs connections to those offices, [it could be the answer]. In that case, the simplest QoS approach is probably all that's required. It's called DiffServ and gives priority to voice.
Weinschenk: So where is QoS being pushed?
Turner: I think there are a lot of people trying to sell QoS solutions. I think there are none whatsoever in the Internet backbone. There is one specific proprietary thing in residential and SOHO [applications], which is to give VoIP priority in DSL or cable modem networks. The point is that the Vonage telephone adapter and the telephone adapters from other services simply provide a priority for voice packets and nothing else. In corporate environments, there are QoS solutions sold. They typically are being employed when different sites have to link over expensive wide area connections. Basically, the preference should be either nothing or the simplest thing you possibly can get away with. Anyone that tries to sell you more than that is trying to sell you something you don't need.