You've asked an open ended question with no specifically correct answer.
To understand bursting and throttling you need to understand where the bandwidth constraints are on a broadband network, and in order to do that, you need to understand how a broadband network is built. Unfortunately this means understanding four common broadband solutions.
The question you've asked is not a technical one really, it's an economic one.
In order to fully detail this I'll have to drop a knowledge bomb - unfortunately it will be disguised as a load of boring old crap bomb and may be rehashing old information that has already been relayed in this forum. For this I apologize.
The four solutions you need to understand are:
LLU (unbundled exchanges with ISP building their own backhaul networks from exchange to POP - DSL circuits purchased through openreach)
IPStream Connect (20C DSL ADSL2 products partial backhaul all provided by BTWholesale)
WBC (21C DSL products with partial backhaul all provided by BTWholesale)
WBMC (21C and 20C DSL products with complete backhaul all provided by BTWholesale)
1. LLU - This solution has the highest fixed cost price, with the least amount of infrastructure purchased by BT
A provider must build out to each individual exchange that they wish to unbundle. This means that each exchange must be furnished with the following things:
At least one DSLAM/MSAN the cost price of these, is anywhere between 10 and 20K per chassis, depending upon supplier discounts
A cabinet of colocation space within the exchange at circa 20k per annum.
A switch or a router of some description (anywhere between 1k and 10k)
Either an SMPF line (if connecting to a DSLAM and just providing broadband services) or an MPF line (full metallic path if the provider wishes to supply broadband and voice services) SMPF line costing ~15 quid per annum and an MPF line costing ~99 quid per annum respectively.
And backhaul connectivity (a gig backhaul product costs between 8k and 15k per annum depending on distance)
In this solution you can figure out roughly what the cost to provide service to a customer is by dividing the amount of fixed costs by the number of ports in the DSLAM/MSAN and whatever the differential is between this and the price you sell a service for is the amount of money you're left with at the end of the month for booze and strippers.
As you can probably already guess from what I've written, the constraint point is the backhaul connectivity. A gig of backhaul for a DSLAM or two may seem like a lot of bandwidth, but when you factor in the fact that most ISPs egress connectivity (the stuff that connects them to the internet) is in london. You also have to assume that remote exchanges such as those in Wales where many ISPs don't have POPs are probably configured in a daisy chain of exchanges. Therefore 3 exchanges with a 1gig connection that are daisy chained from one another only really have about 300mb/s of bandwidth each, between them and the internet.
In this situation, this is where the contention/bursting/constraint resides in an LLU network. Many ISPs will overcome this by putting CDNs/caching closer or in the exchanges, or using DPI to contend the bandwidth. Both of these things are relatively expensive and aren't necessarily practical unless an exchange has a huge serviceable area (more clients). So for the most part, statistical multiplexing is what is utilized in order to provide reasonable service to clients.
Solution 2: IPStream connect. In this instance, as with all the rest, business is done between the ISP and BTWholesale, who will purchase the lines that your broadband is delivered on from openreach themselves, and backhaul the service across their own network to 10 locations around the country.
In each location an ISP must purchase the following items:
MSIL (Multi Service Interconnect Link) which is used to connect from the ISP rack to the BT Node. These come in 1 and 10Gig increments and cost 15k per annum for a 1gig connection and roughly 8x this price for a 10gig connection.
An IPSC interconnect per MSIL (again roughly 15k per annum)
And a backhaul connection, which works in the same way as an LLU backhaul circuit and will cost between 8 + 15k per annum.
So annual costs associated with each IPSC node are roughly 45k/annum. Plus individual DSL line costs.
On top of this, bandwidth is priced at £122.64 per megabit for Markets 2 + 3 customers and £104.22 for Market 1 customers.
In this situation, the contention is different. By far the largest cost associated with delivering your services is the per megabit bandwidth cost from BT. In this situation, CDNs are largely useless for easing of bandwidth constraints, so the only thing that an ISP can resort to is Deep Packet Inspection and throttling to provide a fair service. Because BT bill at 100th percentile you must assume that whatever the ISPs peak traffic utilization will be, is what they'll be billed at for the month. So really as far as end-user-experience is concerned, the times that a user can burst and use as much as they like should be any time that is "off-peak". Peak times, are different, based on individual ISP customer profiles.
A DSL line in this situation costs 6 pounds and (I think 22pence) per month.
So again, you take your peak bandwidth usage, divide it by the number of subscribers you have over the link, divide your hardware and infrastructure costs by the number of customers and if you've got any cash left you can probably afford to buy yourself a bag of chips and a copy of razzle at the end of the month.
Just to make it clear a gig of traffic at ~113 pounds per megabit is 1,356,000 pounds per annum (which is obviously a lot more than the ~8-15k per exchange we were talking about for LLU but if you multiply this number by a few thousand exchanges - it's probably time to hire a mathematician and some really clever accountants to figure out whats best)
WBC - This is the new (As of 2008ish) interconnection solution, similar to IPSC for 21C services. BT have built an entirely new network using MPLS and all manner of cool things, instaed of the old and clunky 20C ATM network. Because routers are faster, hardware is cheaper and links theoretically fatter now a days, the solution is associatively cheaper.
BTW now ask you to interconnect in 20 locations, 10 of which are the locations you already interconnected with IPSC in, which remain as-is for IPStream services. So your 21CN customer base is delivered over all 20 connections, while 20C services remain over the 10 old IPS nodes.
In this situation you still need an MSIL and backhaul circuit as before, but you also need an Aggregation Point per MSIL and an Extension Path per MSIL. These cost 18k per annum and 17k per annum respectively. So your costs are nearer the 60k mark per node
Bandwidth is now reduced however, to 40 pounds per megabit for 21CN customers. While the remaining IPS customers remain at the price mentioned earlier.
Again the constraint point in this solution is your bandwidth across MSIL, where a gig costs more like 500,000 snots per annum. Far cheaper than the 1.3M stated earlier, but still way more expensive than a DIY unbundled solution.
Useless factoid is that a WBC line costs 5 pounds and 80something pence per month if memory serves.
WBMC - is the solution du jour for people that don't have multiple hundreds of thousands of DSL subscribers, but still wish to deliver a broadband solution.
The economics of WBMC are far easier a 1 Gig hostlink costs about 25k per annum and a 10gig hostlink costs about 75k per annum. You plug a router in the end and maybe buy an LNS to terminate your subscribers (an LNS that can happily terminate ~120,000 subs costs between 70 and 100k).
Bandwidth then costs 131 pounds and change for IPStream customers and 48 pounds per megabit for WBC subscribers.
So in this solution when compared against WBC and IPSC your trade off is a much lower amount of fixed costs, and a much more simplified network against a bandwidth increase of between 10 and 20 percent.
Again, in this solution your constraint and contention is your bandwidth cost. Most ISPs in all of these situations will limit the bandwidth that their customers can use through use of policing, shaping and deep packet inspection so that their subscribers never burst above an artificial bandwidth limitation, that they can budget and forecast against.
Reading this you may think "That's a load of old bollocks, BT Wholesale are conning everyone!" *drumroll* "It's far more economic to use build an LLU network and sock BTW off completely"
Good point that man! However, BTWholesale aren't all bad. The unfortunate truth about LLU providers, is that they have far greater flexibility over the networks they build than our old friends at BT.
In order to cover 90% of the demographic populace of the United Kingdom with LLU services, you only need to build out to about 2000 exchanges. Sadly, there are over 5000 exchanges within the UK. BTW provide broadband to all of these exchanges. Meaning that 3000+ exchanges serve 10% of the UK. To add insult to injury, because these exchanges tend to be remote and housed in places that nobody besides the most wayward of sheep are likely to be familiar with, they're also the most costly to deliver backhaul connectivity to.
BTW need to average all of these costs over their overall cost base to provide a product to the common man. All things are equal after all. We live in a politically correct country. Yay socialism!
My numbers might be a little dodgy, particularly the numbers after the decimal place, but you can find all of this information on the BTWholesale and Openreach websites, it's publicly available in their pricelists.
For technical information on bandwidth reservation, management and conservation, it's probably best to pick an individual piece and query that.
Hope that helps.
