RE: The Qos PipeDream [Was: RE: Two Tiered Internet]
Hey there Fergie:
Martin,
You can 'see' anything you'd like, buy your reality does not match everyone else's -- my opinion, of course.
QoS is a myth -- it doesn't exist.
What you're obviosuly trying to tell us is that less-than-best- effort is somehow good? Never sell it.
This vein will come back and bite you guys who think like this.
I'm not sggesting that this be the way the Internet operate at all. The poster asked how this would work if it did (my interpretation) and where there is will (customers) and money (ISP's) there is always a way. The old school in me says "never!", but the experience in me says "possible". I think it *is* unlikely though. Consider the busy signal approach for a second though. Can we build, pay for, and sustain an Internet that never has congestion or is never "busy". If you have a web server and a limited amount of memory or net you tune down the number of httpd's that are spawned and when they are all busy, your site doesn't answer and you get a 404. That's akin to a busy signal and is already in practice today. If I'm Google, for example, I buy thousands of servers so this does not happen. If I'm just plain old me and I am running some popular faq on my personal site, I accept the 404's because I am not going to pay for 100% performance. They can try again later, or, I can pay for more memory or more network to insure optimal performance. Hope that makes a little more sense. And let me turn the question around to you. If the Internet were to work like this, how would we do it?
- ferg
-- "Hannigan, Martin" <hannigan@verisign.com> wrote:
What I'm interested in is how the two service providers will build a two tiered Internet.
The PSTN is tiered both in architecture and operation. Switching hiearchies and a seperate SS7 network which is basically a billing network.
I think the thought is service levels vs. congestion control. For example, CO's have call overflow mechanisms to tandem switch points which basically seek out excess capacity and use it as overflow for call termination if and when possible.
I could see an internet hiearchy where preferred traffic was switch onto hicap overflow links with controlled congestion and other traffic, non premium traffic, "got a fast busy".
-M<
-- "Fergie", a.k.a. Paul Ferguson Engineering Architecture for the Internet fergdawg@netzero.net or fergdawg@sbcglobal.net ferg's tech blog: http://fergdawg.blogspot.com/
Can we build, pay for, and sustain an Internet that never has congestion or is never "busy".
s/never/when there are not multiple serious cuts/ would we build a bank where only some of the customers can get their money back? we're selling delivery of packets at some bandwidth. we should deliver it. otherwise, it's called false advertising. randy
On Wednesday 14 December 2005 23:31, Randy Bush wrote:
would we build a bank where only some of the customers can get their money back?
Not taking into account the FDIC, we already have that, since banks are only required to keep 10% of any given depositor's monies.
we're selling delivery of packets at some bandwidth. we should deliver it. otherwise, it's called false advertising.
No, it's called oversubscription, and it is what produces busy signals/dropped packets/ slow response. What ISP doesn't oversubscribe consumer capacity? When the full cost of that packet at that speed is not passed along to the customer, then the only way for the ISP to remain viable (currently) is to oversubscribe. The same occurs at the telco for POTS. The whole QoS angle is just a way to get people to pay for something they think is better, but is really no different in practice. (Fergie's smoke and mirrors). So in essence there is already a 2 tier Internet, as has been said: consumer-grade (oversubscribed), and real (1:1 bandwidth subscription). Real means that if you buy a T1 or an OC3 or whatever, you get what you paid for, to your ISP's PoP. Consumers don't get this; they get a burst bandwidth at the burst rate, but there is no committed rate for consumers. Otherwise I could get an OC-3 full rate for $1,500 per month (in the boonies, no less). Where problems arise is when those who think they are getting 1:1 real Internet (really just a pipe to their ISP) are actually getting oversubscribed bandwidth instead of 1:1. While marketing seems to be just short of sacrilege to many here, the fact is that NOC personnel salaries have to be paid from somewhere, and if your business is selling bandwidth, then your revenue from customers minus cost of said bandwidth minus operational expenses (salaries, capital, power, etc) had better result in a quantity that is greater than zero, or you're going to be unemployed rather soon. If you are selling $50 6Mb/s DSL, and you're paying $10 per Mb/s, then you have a problem, and oversubscription is your solution. Oversubscribing 4 to 1 makes your non-oversubscribed $240 per month for four subscribers for your bandwidth cost only $60 per month, and you now have $140 per month to pay your NOC personnel and turn a profit (or at least break even). The local ISP here is only oversubscribed two to one; I don't see how they are making any money at all, even with fairly high DSL cost, as I've seen the kind of prices their upstream charges for 1:1 rates. Of course, your upstream (if you have one) also has to make their ends meet, too. At the top SFI level, you still have the cost of transit to worry about, with $1,000 per year or more per mile for fiber maintenance; if you have 25,000 miles of fiber you need to generate $25 million per year to keep it maintained, even if you don't have an upstream. Of course, the $35,000 per mile for fiber installation has to be amortized, too, as does that $2 million backbone router on each end of every hop, etc. Bandwidth has to have a cost; otherwise the bandwidth provider will not stay around long. On the operational end, the challenge becomes designing networks that in the presence of ubiquitous oversubscription degrade gracefully and allow certain features to have lesser degradation. Thus QoS. -- Lamar Owen Director of Information Technology Pisgah Astronomical Research Institute 1 PARI Drive Rosman, NC 28772 (828)862-5554 www.pari.edu
participants (3)
-
Hannigan, Martin
-
Lamar Owen
-
Randy Bush