Conversation on contention ratios

Question 1

Can someone explain to me in a bit more detail about contention ratios; I have just been on the phone to an ADSL provider and I asked "what contention ratio do you offer?", and he explained "none as such, because they are sort of meaningless as there can be different levels of contention between the premises and the street cabinet, and different levels of contention between the cabinet and the exchange, and between exchanges etc". Currently we are with BT and are looking to move away, they offer us a 20:1 contention ratio, is this just false advertising or do they have the power to actually offer this?

Answer 1

BT do not quote contention ratios
BT have not quoted contention ratios for many years.
BT have never quoted contention ratios on 21CN.

It's expressed in terms of "end user experience" - a line is faulty if
you can't download at a certain speed on it, depending on the underlying
product (Higher for Premium products) and the speed the line is syncing
at. I don't have the 21CN figure to hand but as an example, here are the
20CN expected throughputs and below this level is a fault:

BT IPstream Home 250: 50-250kbit/s
BT IPstream Home 500: 100-500kbit/s
BT IPstream Home 1000: 200-1000kbit/s
BT IPstream Home 2000: 400-2000kbit/s
BT IPstream Max, line syncs at 288kb/s: 50 - 250kbit/s
BT IPstream Max, line syncs at up to 576kb/s: 50 - 500kbit/s
BT IPstream Max, line syncs at up to 1152kb/s: 200 - 1000kbit/s
BT IPstream Max, line syncs at up to 2272kb/s: 400 - 2000kbit/s
BT IPstream Max, line syncs at faster than 2272kb/s: 600* - 7150kbit/s

BT IPstream Office 500: 200-500kbit/s
BT IPstream Office 1000: 400-1000kbit/s
BT IPstream Office 2000: 800-2000kbit/s
BT IPstream Max Premium, line syncs at 288kb/s: 100 - 250kbit/s
BT IPstream Max Premium, line syncs at up to 576kb/s: 100 - 500kbit/s
BT IPstream Max Premium, line syncs at up to 1152kb/s: 400 - 1000kbit/s
BT IPstream Max Premium, line syncs at up to 2272kb/s: 800 - 2000kbit/s
BT IPstream Max Premium, line syncs at faster than 2272kb/s: 1200* -
7150kbit/s
* 90% of the time.

This makes sense because you know what you're getting, you don't have to
take a providers word for it on contention ratios and you don't suffer
just because of a few heavy users on your exchange. Many LLU providers
quote 20:1 and 50:1 as being equivalent to the BT products. Whilst it's
true that BT used to quote these figures, they abolished them as in
practice they were doing much better than this: Typically 7:1 and 12:
worst case figures for home and office products respectively.

Answer 2

He's sort of right, BT Wholesale moved away from using contention ratios a number of years ago as they weren't implementing them...

What they now offer is 'wholesale experience metric' which says that for X amount of the time you should expect to receive at least Y bps (with X and Y depending on the product).

If you performance cosistantly drops below Z then you can raise a fault (where Z is a fraction of the maximum stable rate that BT Wholesale has established is correct for your line)

On top of this the ISP has to do their own capacity planning to manage their central capacity, which is a bit easier today as they can install a 1Gbps connection and then flex the bandwidth up and down (albeit on a minimum 3 month term)

Answer 3

Just to try and explain - "contention ratio" is a term used to define the ratio of theoretical maximum use of a link and its size. So for a simple example if you had, say, 20 customers with 500K lines to an exchange that then had 500K back-haul that is 20:1 contention ratio. I.e. there are 20 times the total possible usage demands as capacity.

The reason this comes up is that when BT first launched broadband links they had fixed speed services (500K, 1M, 2M) and stated contention ratios (which is basically stating their planning rules). They operated two types of ATM VP, one at 20:1 and one at 50:1.

There are a lot of things wrong with contention ratios as a metric. For a start, there is a huge difference between 20:1, 200:10, and 2000:100 in practice but they are all a ratio of 20:1. For example, 2:1 is considered a "good" contention ratio - but if you and your neighbour have a 1M line and you both share a 1M backhaul - *any* usage your neighbour makes reduces your usage. But if there are 100 of you with 1M lines and a 10M backhaul (10:1, which seems a lot worse than 2:1) then it takes more than 9 neighbours worth of traffic running flat out to take anything away from your usage. That is a lot less likely. So ratio alone is not meaningful.

Another big problem with contention ratio is that it is irrelevant as anything more than a planning rule. Regardless of the ratio (with the exception of 1:1), if you have contention then you could have slower service because of other peoples usage - simple as that. The ratio is not a huge clue as to how likely that is because it depends on the other people, and internet usage can vary massively. With the right people (light usage users that check email and facebook occasionally) a ratio of 200:1 may be no problem at all as the links do not get full. With a few bit torrent users then a ratio of 2:1 (or 20:10 or 200:100) may not be good enough to avoid congestion.

At the end of the day - if the shared pipe is not actually full then the link you have is as good as a 1:1 un-contended link. If it gets full then it is congested and that means problems regardless of "contention ratio". Without knowing the actual usage patterns you cannot tell if a specific contention ratio will mean congestion or not.

This means that even when BT sold 20:1 and 50:1 links, that was not itself a clue which was better (i.e. less likely to be congested) as it depends what sort of people bought those links and how much they used. If all the people buying 50:1 actually averaged 1/100th of what they could, but all the people buying 20:1 tried to average 1/19th of what they could use then 50:1 would be "better" than 20:1...

Just as a further example, if your computer has a 100M LAN and the router has 100M port, and you have one PC, that is 1:1 from you to router. Router has, say, 10M ADSL. That is a 10:1 ratio, or 1:10 ratio depending on which way you are going, but as you are the only user that does not matter. The link to the exchange is exclusive to you, so that 10M is 1:1 all the way to the BRAS/MSAN in the exchange. The exchange to the BRAS will have a contention ratio which could in theory be worked out based on total users and total back-haul. The BRAS then has links over the core MPLS network and gets to ISP interconnect - that ratio will be a pig to work out as there is not likely to be a single pinch-point. BT do get those links congested at times though. The link to the ISP is probably the biggest congestion point for most broadband users.

Then you get to "the internet". Imagine a web site you like, maybe it is an on-line shop that runs on *a* server, and lets say it has a gig port with a whole gig of transit available. In practice that gig would be shared with other hosting customers, but assume for a moment that on-line shop has a really good gig link to "the internet"... Now imagine the total possible usage of every single internet link everywhere in the world that could, in theory, connect to that on-line shop... I have no idea what that works out to, but it almost certainly means that to that point on the internet you have many millions to one contention ratio.

As you can see - it is a pretty meaningless metric even if you can measure or calculate it.

Hope that helps.

 

Question 2

Just to throw a stick in the mud;

http://business.bt.com/broadband-and-internet/internet-access/sdsl/

BT are claiming a contention ratio here but is this simply because its SDSL so they are using fixed throughputs and dedicated back-haul for their SDSL possibly?

Answer 1

No, because its a forgotten product and they haven't gotten round to updating the copy. BT SDSL is normally delivered over IPStream (although there are other alternative) centrals and backhaul network so its capabilities should have been restated at the same time as ADSL

---