The announcement by Safaricom that it’s doing away with its unlimited Internet bundle did not come as a surprise to me. I had discussed the historical reason behind the billing model that is used by ISP’s and mobile operators in a previous blog post here in Feb 2011.
The billing model used in unlimited Internet offering is flawed. This is because the unit of billing is not a valid and quantifiable measure of consumption of service. An ISP or mobile operator charging a customer a flat fee for a size of Internet pipe (measured in Kbps) is equivalent to a water utility company charging you based on the radius of the pipe coming into your house and not the quantity of water you consume (download) or sewerage released (upload).
What will happen if the local water company billed users by a flat rate fee based on per-centimeter radius of pipe going into their homes rather than volume of water consumed? A user with a pipe of radius that is 1% more than the neighbor enjoys 2% more water flow into their house (do the math!). The problem is that their bills will not differ by 2% but by 1% based on the difference in radius of the pipes. A 2% difference yields a 4% difference in consumption but a 2% difference in billing. The result is that a small group of about 1% users end up consuming about 70% of all the water. This figure is arrived at as follows: A marginal unit increase in resource leads to a near doubling of marginal utility. This is a logarithmic gain (Ln 2=0.693 which means that 69% of utility is enjoyed by about 1% of consumers) . This is the figure issued by Bob Collymore the CEO of Safaricom who said that 1% of unlimited users are consuming about 70% of the resources. This essentially means costs could outstrip revenues by 70:1. This does not make any business sense. Not even a hypothetical NGO engaged in giving ‘free’ Internet through donor funding can carry such a cost to revenue ratio. As to why ISP’s and mobile operators thought billing by size of pipe to the Internet could make money is beyond me.
Bandwidth Consumption Is Not Linear
One mistake that network engineers make is to assume that a 512Kbps user will consume double what a 256Kbps user does and therefore advice the billing team that billing the 512Kbps twice the price of the 256Kbps can cover all costs. This is not true. There are things or activities that a 256Kbps user will not be able to do online, like comfortably do Youtube videos. A 512Kbps user will however be able to do Youtube without a problem. The result is that a 512Kbps user will do much more Youtube videos as the 256Kbps user becomes more frustrated with all the buffering and stops all together attempting to watch online videos. The result is that the consumption of the 512Kbps user will be much higher than double that of the 256Kbps user. Other than Youtube, websites can detect your link speed and present differentiated rich content based on that. I’m sure some of us have been given an option to load a ‘basic’ version of Gmail when it detects a slow link. The big pipe guy never gets to be asked if he can load lighter web pages, rich content is downloaded to his browser by default while the smaller pipe guy gets less content downloaded to his browser in as much that they are both connected to the same website. The problem here is that the difference in content downloaded by the two people on 512K and 256K link is not linear or even double but takes a more logarithmic shape.
Nature Of Contention: Its a Transport and not Network problem
The second mistake that the network engineers make in a network is to assume that if you put a group of customers in a very fat IP pipe and let them fight it out for speeds based on an IP based QoS mechanism is that with time each customer will get a fair chance of getting some bandwidth out of the pool. The problem is that nearly all network QoS equipment characterize a TCP flow as a host-to-host (H2H) connection and not a port-to-port (P2P not to be confused with Peer2Peer) connection. There could be two users with one H2H connection each but one of them might posses about 3000 P2P flows. The problem here is that bandwidth is consumed by the P2P flows and not the H2H flows. User with the 3000 P2P flows ends up taking up most of the bandwidth. This explains why peer to peer (which establishes thousands of P2P flows) is a real bandwidth hog.
So what happens when an ISP dumps the angelic you in a pipe with other malevolent users who are doing peer to peer traffic such as bit-torrent? They will hog up all the bandwidth and the equipment and policies set will not be able to ensure fair allocation of bandwidth to all users including you. So some few users doing bit-torrent end up enjoying massive amounts of bandwidth while the rest doing normal browsing suffer. That explains why some users on the Safaricom Network could download over 35GB of data per week as per comments by Bob Collymore. Please read more on how TCP H2H and P2P flows work here. Many ISP’s engage engineers proficient in layer3 operations (CCNP’s, CCIP’s, CCIE’s etc ) to provide expertise on a layer 4 issue of TCP H2H and P2P flows. You cannot control TCP flows by using layer 3 techniques. IP Network engineers are assigned the duties of transport engineers.
At the end of the day, there will be a very small fraction of ‘happy’ customers and a large group of dissatisfied and angry customers. The few happy customers flat rate revenues are not able to cover all costs as the unhappy customers churn. If on the other hand these bandwidth hogs paid by the GB, the story would be very different. This is what operators are realizing now and moving with speed to implement. Safaricom is not the only one affected by this; Verizon, AT&T, T-Mobile in the US are all in different stages of doing away with unlimited service due to their unprofitable nature.