[Ignitionnet]

Responses inline:

Originally Posted by Chrysalis

your 2 examples.

BT shape certian protocols to insanely low speeds, so it only works when you throttle users down to sub 200kbit on things like p2p.

TalkTalk also traffic shape which skews their figures but are upgrading to increase their allocation per customer 10 fold and are rumoured to remove traffic management later this year for their LLU customers.

Talk Talk prided themselves on how little bandwidth they could have their customers consuming. The upgrade is due to them offering ADSL2+ and uncapped services and will still include traffic management.

I agree that both shape - which is why Virgin budget about 4 times as much bandwidth as those guys.

The cable isp's in america by 2012 plan to allocate 1mbit per end user.

I would be interested in seeing some kind of evidence of this. I find it somewhat unlikely that they'll be allocating a card for each 76 customers, or in the case of DOCSIS 3 a 4 downstream port covering 152 customers.

I keep a fairly close eye on the US cable industry along with reading industry magazines and can't find anything suggesting that they intend on going that low - it would require total elimination of all intermediate coaxial amplifiers and for all cablecos to be using a fibre deep solution.

But the key thing here is contention works far better when the shared pool of bandwidth is larger.

Statistical contention. Quite agree.

so 100 users sharing 100mbit is far better than 10 users sharing 10mbit even tho both are the same contention ratio. VM's shared pool of bandwidth is split into very tiny portions so requires a lower contention ratio to handle moderate to heavy use. Entanet are a good example of this, they were able to originally share their entire BT central bandwidth across their entire national customer base on ipstream, when they moved to 21CN WBC they then had fragmented their customer base into smaller groups, this led to problem areas (sound familiar?) some of their nodes which had a good balance of light residental users vs business customers were ok, others which were no longer able to be subsidised by the non residental customers struggled with much lower performance and more congestion. When bulldog tried using small shared backhaul on datastream (only 2x the end user's burst) it was a disaster.

Entanet's WBC experiences aren't a case of statistical contention but of load balancing. Exchanges being statically mapped caused their issues, not access speed versus backhaul speed. 22Mbit out of 1Gbit is still nothing - look at LLU.

Bulldog's disaster was in no small part due to selling 2Mbps on 2Mbps. Not clever. There were plenty of VPs that were sized at 4Mbps with 2Mbps customers on them running just fine. There were also plenty that weren't, all down to how they are used.

I think we will disagree on this point here, my judgement on VM is how it deals with its worst off customers stuck in a ring fenced node with severe congestion whilst other isp's are able to have a fairly equal congestion applied across their entire customer base so they all suffer in tandem sharing the pain. The evil variance on adsl instead is line lengths.

All nodes / service groups are ring fenced. This is the nature of cable for better or worse.

We don't disagree on anything, I'm not saying VM or anyone else is right or wrong, just that this is what they do and these are the thought processes. As you may remember I've worked for a couple of ISPs here and there

When you pointed out to me in another thread VM are able to mitigate congestion by going down to around 15:1 contention this said it all for me, especially when NTL were quoting 20:1 in their TOS some years back. So at one point they used to contend around that level and clearly since then corners have been cut and overselling increased. I feel you making the mistake so many isps make in that you assuming as burst speeds go up usage will not go up with it, and assume that only a tiny proportion of users use their connection moderatly, I disagree on that, the users who only browse a few websites and read email are shrinking rapidly.

Not really - still a number of people browse and email, average usage is still less than 10GB/month from Joe Average even with You Tube and iPlayer.

Contention ratio did indeed increase, why do you think STM came about?

ISPs don't make 'mistakes' with regards to the bandwidth consumption of their customers they have comprehensive statistics on how much bandwidth is being consumed on their networks. They also have budgets to stick to so have a balancing act between the two. In a perfect world every ISP could go cap in hand to a sugar daddy such as Telefonica and get a few million quid to upgrade their network but in the real world the money men still hold the purse strings that the engineers need to get at.

There is no target contention ratio that any ISP has now, beyond to go as high as possible without getting too many complaints. That is really how VM / ntl / TW have always operated regardless of any advertised contention ratios that they might happen to have on the way. If a node is outside planning guidelines it gets fixed, if a node is overutilised but is within guidelines it gets thought about a little more.

But congrats to VM, by ring fencing off problem areas they also in affect hiding the weakness of their infrastructure to more parts of their customer base so the problem appears less severe than it is. The fact they wont move users of my port to another highlights this, as it would then probably create congestion for other users who dont see it now.

As I said it's not about ring fencing. To move customers to other ports, with the exception of the DOCSIS 3 network, requires physical work to split the node. Cable is a ring fenced network, segmented by physical fibres.

If the DOCSIS 3 network is in strife there's not really much of an incentive to make that network worse by pushing people across to it - I don't know what the situation is with that.