The Internet bandwidth problem in Malta

We keep hearing subscribers complaining about the poor service and at the same time the ISPs are complaining that subscribers are 'abusing' the service. To me it appears that the cause of the problem is obvious, and even more so its solution.

ISPs have historically charged a fixed monthly rate independent of usage. This created massive problems already in the days when dialup was the only option, since everyone and his brother would dial up exactly at 6pm (when the phone rate becomes cheap) and leave the PC connected regardless of whether they are using it or not. This created a bottleneck in terms of the number of simultaneous connections the ISP could serve, in other words how many modems and phone lines the ISP had installed. The ISPs took a considerable length of time to learn and started to change the way they charged by having the fixed charge cover a bundle of hours and then charge for any extra hours of use made. This solved the line availability problem almost overnight.

Along came ADSL with a whole lot of promises of miraculous things, we could watch high quality streaming video, we could download unlimited stuff very quickly and all sorts of stuff. The first few people got ADSL and it was 'wow'. A few more people got it and it was OK, more people got it and it is now dead slow. The subscribers are moaning because it is slow and the ISPs are moaning because their customers are complaining.

Funny thing is that only a tiny minority of ISPs learnt from the past mistakes with fixed rates on dialup access. Many people got their ADSLs installed and when not doing anything particular either leave some video stream running even if they are not watching it, or leave it permanently downloading stuff from Kazaa and the likes, probably discarding most of the downloaded files because they were not needed anyway. Problem is if they are not paying for the extra download there is no incentive to conserve bandwidth. Everybody uses (wastes) as much as they can and nobody gets a decent service.

ISPs have introduced all sorts of measures for limiting bandwidth, but to me it all seems pretty pointless. Why have a fast modem if there is nothing to feed it with?

It is often claimed that 10% of ADSL users are consuming 90% of the available bandwidth. What  I am proposing is a 'clean' method to rebalance this to some extent, as well as rebalancing charges to make sure that heavy users contribute more to strengthening the infrastructure than the light users.


So what is the solution?

Quite obvious, if 10% are using up 90% of the resources, make sure that those ten percent pay for 90% of the costs AND use the extra money to lease more bandwidth.

Of course one can expect that once subscribers are paying on a per volume basis they will be more careful with their usage. Those who still want to download huge quantities may still do so, but they will pay more cash, which will be used by the ISP to get more bandwidth.

For some strange reason, many ISPs turn white when one mentions per volume charging to them, as if it were some sort of sacrilege. I really can''t figure out what the problem is - it is certainly not a technical one.


Some numbers

So lets have a look at some numbers to get a clearer understanding of where the money is going and what it is doing. The numbers are not accurate at all, but should be somewhere within sight of reality. I have made a lot of assumptions, which may not be accurate, but I have stated them all, so if one has more accurate information it can be slotted in to get a more accurate result.

It is of course a very simplistic calculation, but its purpose is only to illustrate an idea. Who knows maybe it will even inspire someone to do the calculations properly and get the ISPs to listen.


What are the costs faced by an ISP ?

The costs of an ISP will consist of various things. I have listed what I believe are the major costs,  together with my blind guess of what it costs per month

  1. An overall overheads cost which will be roughly proportional to the number of subscribers, lets say LM 1 per subscriber
  2. A recurrent cost for paying Datastream for the ADSL connections, lets say LM 5 per subscriber
  3. A recurrent cost for paying Maltacom (or is it Datastream) for the international 2Mb/s links, which I believe is about LM2000 per Mb/s

The average subscription rate for a 256K ADSL connection is about LM13. So from the figures above, LM6 goes for items 1 and 2, leaving LM 7 for item 3 and some profit, lets say LM2 per subscriber per month.

That leaves LM 5 to pay for the international connections. (in item 3 I am including not only the transport cost but also the overseas termination and connection into the wider internet world)

A simple calculation tells us that to run the business at some decent profit, an ISP will need 400 subscribers to finance each megabit/s of bandwidth, or put another way 800 subscribers for each 2 megabit link since to my knowledge capacity can only be purchased in chunks of 2 megabits per second.

Under this scenario, let us for a moment assume that all 800 subscribers leave Kazaa running 24 hours a day downloading stuff continuously. How much bandwidth will each have?

This works out to a pretty dismal 2.6 kilobits per second, barely faster than the modems we had when internet was first introduced here in Malta.

Now obviously not all subscribers are downloading 24/7, but certainly a substantial number of them are. 


The solution

I would suggest a tariff structure similar to that used by most cellphone operators for their contract subscribers, but with some differences as well. This would consist of these characteristics:

  1. A fixed monthly charge
  2. A bundle of 'free' international volume
  3. A per megabyte charge on international traffic once the bundled volume is used up
  4. Two or possibly three charging rates at different times of day for the applicable volume charges
  5. The time window for the bundle is staggered.


1. The monthly rate

Since there will always be people who will exceed the bundled volume allowance, and therefore generate additional revenue, it should be possible to reduce the fixed monthly rate a little, lets say from LM 13 to LM 12 or perhaps even LM 10.


2. How big should the 'bundle' be

Since we know that people in general are greedy, and will most likely make sure they squeeze out the very last bit of their 'free' bundle, we must make sure that given this factor there will be a respectable average throughput for subscribers in general. 

A 2 megabit per second link can transfer a total of about 500 Gigabytes per month. Dividing that by our 800 subscribers would leave us with about 600 megabytes per subscriber. This figure is a first approximation of what the bundle size should be.

Since there will be some who exceed this amount and others who use much less, we can raise the bundle size to about 1 Gigabyte (even if just to have a round number).

Thus if all the subscribers were to try to download more or less this amount of data, the link would be at about 100% utilisation and may be sufficient. There are of course many other factors to consider, and my gut feeling is that the bundle should be closer to 500MB to avoid problems (in which caser the monthly rate would be reduced to, say, LM 10)


3. How much to charge per megabyte

This is  a relatively easy one. We have seen that a leased line that can transfer 500 GB per month costs around LM4000 per month, meaning that all considered a gigabyte costs about LM 7


4. Different rates at different times of day

Time bands are an excellent way of maximizing the utilization of just about any costly shared resource. As I said before people are in general greedy. The only way to control their usage effectively is by adjusting the cost.

If possible we want to keep the links as free as possible from massive downloads at times when people are most likely to require interactive use. Most probably the peak time for interactive users would be from 5pm to 11pm. We can make this the most expensive time band (for volume over and above the bundle). Most inetractive users would not exceed the bundle limit anyway, so they will not incur any extra cost. 

The heavy downloaders on the other hand will have a very powerful incentive to do their downloading at other times. The reasons why I propose three bands rather than two is as follows.

Lets assume rates of LM 12, LM 5 and LM 2 respectively for the Peak, Normal and Cheap rates respectively.

The LM 12 rate is intended to downright scare off heavy downloaders from this time band, which would probably be 5 Pm to 11 PM.

The LM 5 rate will be a good incentive to avoid downloading useless stuff when people are at work using the internet for useful things. This time band would probably be from 7 AM to 5PM and 11 pm to 2 am.

The LM 2 rate is intended to actually attract down loaders to the time band, which would otherwise be unutilized by both home and business interactive users.  It would probably become crammed and congested at these times since everybody would be downloading at this time. But at least it would impact much fewer customers than it does today. The time band would probably be between 2 am and 7 am.

Of course the real rates and time bands would need to be worked out on a more scientific basis, but as I said before, here I am trying to illustrate a principle and a structure rather than accurately mapping the economics.


5. Why a staggered time window?

This is an easy one.  The behavior of people with 'bundles' is quite predictable. Usually one will be a bit careless as soon as the bundle starts, because he suddenly has a lot of free megabytes. About 30% of the way through the time window he will start worrying that the bundle won't last to the end of the time window and start being over cautious with consumption. Then towards the end of the time window he suddenly realizes that he has a lot of free megabytes which he will loose if he does not use up quickly. More likely than not he will use up the remaining amount even if he just wastes it.

So, the problem is that if everybody behaves the same way at the same time we end up having a network which is always congested at the beginning and the end of each month and is grossly under utilized in the intervening period.


What does it mean to the subscriber

Subscribers who do not 'overuse' the net currently will probably benefit from a slight reduction in the monthly rate. I previously guessed this as a reduction from LM 13 to LM 11. They will also benefit from much faster response time as well as download times.

The heavy down loaders will most probably download much less than they used to, but will probably be more specific on what they choose to download rather than just downloading anything and thrashing 90% of what they downloaded.

Assuming that they stick to the cheap time bands, they will not be faced with spectacular cost either. To give a few examples:

A whole hour of MP3 music would cost about 10 cents. 

Watching streaming video at 128Kb/s would cost about 10 c per hour

Watching streaming video at 512Kb/s would cost about 40c per hour

One full length VCD would cost about LM 1

A full quality DVD would cost about LM 10

The average PC game would cost about 30 cents

(Of course in all of the above I am only speaking of the costs payable to the ISP, not the cost payable to the content provider where applicable)

The cost of downloading a DVD seems exorbitant, but I don't think downloading DVDs from the Internet really makes sense anyway.

Of these, the only things which would conceivably be done during peak time would be watching video streaming (or listening to streaming audio), in which case the costs would be of  about LM 2.40 per hour for 512K video and 40c per hour for 128K video. Sounds extremely expensive, but you have to consider that today it is not even possible to stream at this rate, as even 64kb/s streaming rarely works properly due to the massive network congestion (and compounded by the brain-dead bandwidth limitation policies currently being used)



There is no conclusion actually - this article is meant just as 'food for thought'. I don't claim to know any more than ISPs do, but hopefully someone will read it and maybe some of the ideas are useful.