Paul Kapustka's Blog
« Telecom money keeps flowin' to Congress | Main | One (smart) definition of network neutrality »
May 16, 2006
Why QoS sucks... money out of your pocket
As some of my critics have noted, I'm not a network engineer. But somebody who is just penned a long, convincing essay as to why QoS isn't the answer for better bandwidth, but is instead mainly a telco ploy to suck more cash out of users while keeping alive their old, tired networks.
It's a long read, but extremely worthwhile. One of the many money quotes says:
The operator controlled QoS challenges and associated expense dramatically increase when you consider providing consistent QoS behavior, via policy, across multiple technologies, service providers, applications and devices. It is delusional to think that network service providers will share their policy information which they all view as a competitive advantage. This leads to the conclusion that best effort is fundamental for competitive behavior and makes end to end QoS an unreachable goal.
And, further:
There is also a clear danger that the cost of the QoS mechanisms needed, including their operation and management will far exceed what is required to provide ample non-QoS service, or simply bandwidth.
And for critics of critics of the "need" for QoS:
Attacks on the anti-QoS perspective often center on the claim that the anti-QoS crowd want “infinite bandwidth". This and similar statements by the QoS advocates are nonsense! Adding more bandwidth has always been more cost effective than QoS. Others also agree, please see http://www.dtc.umn.edu/~odlyzko/doc/internet.economics.pdf. Think about this, if adding bandwidth wasn't effective, then how come QoS isn’t widely deployed today? A look at QoS, past the hype, reveals numerous issues. As an Internet user, all I care about is that I have enough capacity so most things run nicely. Just give me my advertised bit rate and I am happy.
(What's really funny is that the same crank who tried to tout the wonders of QoS in previous comments on my blog has also commented on the link where all this info came from. The truth hurts, but apparently not enough to convince everyone.)
Please read, and decide for yourself!
Thanks to Martin "no net neutrality for me, just more Internets" Geddes for the link.
Posted by paul at May 16, 2006 09:17 PM
Comments
The author cited seems to be under the impression that end-to-end QoS requires competing network providers to share policy information. That's nonsense. All it requires is a standard for marking delay-insensitive packets differently from delay-sensitive packets. Who cares what's in the packets from another network as long as you know that someone's paid for QoS treatment? Differential pricing is all that's needed to make this work, not any sharing of policy information.
The author is correct that you can guarantee QoS by ensuring an abundance of bandwidth. But it's pretty clear that even in fairly competitive markets (like the market for enterprise network services) at some point QoS is cheaper than bandwidth. It is a total fantasy to think that it is cheaper to eliminate oversubscription in a network than it is to introduce QoS.
QoS is in fact widely deployed today in networks that are shared by delay-sensitive and delay-insensitive packets. The author is simply wrong on this point.
Posted by: d.l. at May 18, 2006 11:26 AM
"This leads to the conclusion that best effort is fundamental for competitive behavior and makes end to end QoS an unreachable goal."
Tell that to an IT department at a big business. They will get a kick out of it since they are already paying for and receiving QoS.
Posted by: MnZ at June 11, 2006 05:45 PM
The previous comment is from someone who has never worked on a network and uninformed about network economics. Policy is important for QoS to work network wide. The clueless individual from the marketing department ...oooh look, just mark a few packets and QoS is done. Talk about over simplification ...pathetic! Also, try installing QoS on a network wide basis. It is nothing but a huge expense and more bandwidth has always been cheaper. Moore's law give us increasing capacity at lower costs and people to manage this stuff as one small piece keep getting more expensive. If QoS was so wonderful and cheap how come it isn't widely deployed.
Posted by: Mr Bandwidth at November 13, 2007 09:43 AM