Some calculations that dont add up

[Daveogg]

So how is this gonna work???

Accepted contension ratio ~ 50

Therefore with 4000 users iburst will need 80MB/s of bandwidth. This fits with the prelaunch usage of 20 - 60 MB/s dynamically adjusted.

If they contend it greater than that we will "feel" the effects of that as in the situation for the past week.

However at 80MB/s there is a total of 207360 GB/ month available.

Now each user is capped at 3GB so only 12000GB are actually needed.

Ie to maintain a decent contension ratio so that we all have speed when we need it iburst has to provide 15 times the cap per user.

 

[guest2013-1]

no not really, only if all 4000 users are busy downloading at full speeds (and uploading at full speeds) will they feel they effects.

Maybe someone like TheRodent can answer this?

 

[swordfish1]

see, it is the monopoly that makes this 3 GB limit! Because telkom has 3 GB limit, so iBurst decided to screw us and make 3 GB limit, and charge us more if we want extra ... we don't have an option!

The rest of the world either does not have a monthly limits at all, ot they start from 20 GB upwards. I don't think it is so much the lack of bandwith the problem, as it is the opportunity to make more money. There are a lot of rich people with lots of money in SA that don't mind paying few thousands rands just to get let say 20 GB per month. They will make much more money that way, rather than saying ok lets give everyone 20 GB, then no one will buy any extra bandwith.

So it all comes down to the lack of competition. If there were 10 independent companies offering similar service (not reseling each others services!) , I bet the monthly limits will start at something like 20-30 GB per month if there are any ...

btw I just come accross internet prices in another third-world country - the dial-up is free, 2 Mbit cost US$40 per month with no limits whatsoever .... how is that for comparison!

 

[DFantom]

If you have 4000 users on the network, you make sure your network can only handle 75% (3000) of them running at 80% usage since statistically thats what you get. There is not one ISP that doesn't work like this, although the numbers may vary slightly.

There is also a lot of factors that make up bandwidth for iBurst. There is tower bandwidth, so how much the tower has to connect to there network centers, then their backbone bandwidth to the main center, then they have their bandwidth to UUNet for local and international. I have made some assumptions of how the system is configured, but generally it is more complex than what you are working on.

 

[Johand]

noone, The fact that not all the users are downloading at full speed is captured in the contention ratio feature.

The calculations does not add up, because I doubt that they will get an 80 MB/s connection (i.e. 640 Mb/s connection). They run over uunet (you can check it by doing traceroutes etc) which doesn't have that capacity - please see a description of their network at:
http://www.uunet.co.za/_about/network/international/default.asp?n=937
Thereis a total of 180+45 Mbps CIR as far as I can see. This description might be outdated, but there is definitely not 640Mbits/s bandwidth on UUNET in total, nevermind the capacity to sell that kind of bandwidth to iBurst.

You also have to remember, that subscribers do get a 64kbps connection after the cap to the internet, which you will have work into your figures.

And finally, I do have doubts whether a 50-to-1 contention ratio will do it for broadband, always on users.

 

[DFantom]

The calculations does not add up

they do for an ISP based of statistical math. Stop using linear maths and think.

but there is definitely not 640Mbits/s bandwidth on UUNET in total

Agreed. In fact I doubt there is that much intl bandwidth for the whole of South Africa, let alone UUNet. But once again, there is more than 1 type of bandwidth.

swordfish1, is very right IMHO. The 3Gb was chosen because ADSL default cap is 3Gb.

 

[guest2013-1]

Johand, this is where transparent proxying comes into play etc etc etc.

Theorettically speaking, thats what they need, but thats not what they require. Mweb now has a couple of hundred THOUSAND users, and still they're using less than 60meg. It's true that most of those users are dialup customers, but if all dialup customers come online after 7, how are they able to work on such little bandwidth?

Because not everyone downloads all the time, those 2 seconds you take to click on a link, someone else is using the bandwidth... etc etc

Contention ratios are just the tip of the iceberg that makes up the entire ISP game

 

[Johand]

I was using linear maths, but the case is even worse if you do proper statistical anaylis. In that case, their math must be something like: We can accomodate 75% of subscribers at 80% at peak usage times.

The problem is that a cap does not make people use the network efficiently - access to the network will form a gaussian distribution, and subscribers will perceive service as bad when peak cases are not handled to a degree. This is the reason why voice operators (cellphone or landline) do two things:

1) Allocate free minutes to users evenly through the course of the month. This is to prevent everybody calling on the same day (when they receive their minutes)
2) Cellphone packages are cleverly designed to balance load on the network - (i.e. not only "use the network on weekend", but try to use the network for that much over weekends and that much during weekdays)

So my response is: yes, i've not done proper statistical analysis, and my reply is, lets do proper statisical analysis. DFantom, the statics you mentioned, is in fact only linear math - multiplying percentages.

As to the suggestion that transparent proxies etc. will really help I can't see how it will:
1) Proxies does have a limited effect on peer-to-peer
2) Proxies are useless for streaming media
3) Proxies are useless for online gaming
4) Proxies are useless for e-mail
5) Proxies are (almost) useless for websites that generates content dynamically

You do not use 3GB by surfing the web (large http/ftp transfers excluded, and I'm not certain how much of that is cached)

Of course local access does make a difference, but if international bandwidth is not that important, can somebody please explain to me why ADSL users are kicking and screaming when they reach their cap?

MWeb can get away with higher contention ratios because
1) Users pay for the time they spend online, their usage time is shorter
2) Because of callmore, their users are more evenly distributed in time
3) Dialup users use the internet more for e-mail, html etc. of which html is cached, instead of long running peer-to-peer, streaming media and large downloads.
4) They already have contention ratio's on the number of incoming lines

Broadband users have a 24/7 connection, and thus are using it when it is most convenient for them. The problem is that you get a lot of peak demand times with this scenario.

If we would like to do this proper statistical analysis we need to have more information:
1) What is the average amount of data that users transfer a day
2) What is the standard deviation of this mean
3) What is the usage patterns time-wise (hour-to-hour)
4) What is the usage patten during the month (do you go for your cap in the first day, the last day or spread it evenly)
5) What is the ratio between local/international access
6) what is the cacheable transfers

I would also like to note that some broadband users expect a leased line - which, for obvious reasons is ridiculous.

 

[ic]

Of course local access does make a difference, but if international bandwidth is not that important, can somebody please explain to me why ADSL users are kicking and screaming when they reach their cap?

Well IMO that's bcos Telkomonopoly advertise that ADSL customers still have International access once capped, that such access is speed/throttle limited, but that it is usable. The reason why ADSL customer complain is bcos Telkomonopoly are lying to them - it is practically impossible to even smurf international pages when capped - I don't have ADSL myself, but I have observed this with someone that is an ADSL customer of Telkomonopoly.

It's a simple case of not getting what is being paid for, and despite ASASA.org.za's dumbass conclusions to the contrary, it is false advertising on Telkomonopoly's part.

Anyways, that's a discussion that has been ongoing now for ages in the ADSL forums on MyADSL, and also what concerns me about WBS's 64kbits/s throttle when capped - is it going to be a steady minimum of 64kbits/s [even just for local] or is it going to be 64kbits/s contended to the point where it's extremely unreliable & just a method of getting iBurst customers to purchase additional bandwidth & re-enter the vicious money spending cycle to feed their speed addiction...

 

[LoneGunman]

if IBurst have mistakenly banked on the idea that their main profit will come from the purchasing of 'extra' bandwidth, then its going to be real funny watching Sassan and the others faces grow greyer as the months go by and more and more and more customers purchase less and less and less 'extra' bandwidth, and make life increasingly like hell for the Helpdesk, and continue downloading everything they can, 24/7.