Tiscali and its network problems
In the last few weeks there has been an upsurge in the number of posters complaining of the speed of the ADSL services that Tiscali offer. Sites like Web-User featured the news that Tiscali are promising that upgrades between the 27th March 2004 and 7th April 2004 should solve most of the problems.
Tiscali is a large user of the BT Datastream with In-Span Handover system, whereby a teleco can rent dedicated backhaul from the exchange to the teleco's own network. This can offer cost savings, and is also how Tiscali are able to offer their 150kbps and metered products. BT Datastream products have to date had a mixed reception from users of high contention/low price services, this is in part due to the costs and therefore attractiveness to ISPs to use as small amount of backhaul as possible, as it is to the low prices attracting users who will want more for less.
During the last year, there have being a number of posters who have spoken with Tiscali support, and support has informed the user that they have being earmarked as a heavy user and therefore their connection will have a lower priority at peak times. Some see this as throttling users, others call it capping, which strictly speaking it is not. Since the throttled user is actually only slow until their specific exchange has spare bandwidth, which may be five minutes or five hours away.
Looking into this 'throttling' by Tiscali, the methods employed have become clearer, and it works something like this. Imagine an exchange with 100 Tiscali 0.5Mbps consumers, they are sharing a 2Mbps backhaul (often called 'Virtual Path' in the forums). This means that only 4 users can download at their maximum speed, if 20 users are downloading from a sufficiently fast enough source, the speed will average out at around 100kbps. Back in 2000 when ADSL was first launched in the UK, Internet traffic was bursty, but increasingly people are using bandwidth in a constant manner. This can catch ISPs and even BT Wholesale out at times, as often network upgrades are based on projections from past years.
Of course not all users are using the Internet all the time, but in our example, even if just 10 people are using it a lot for P2P and downloads then the performance for others is impacted drastically. A number of solutions exist, one is to try and acquire more customers who are not heavy users and use the new revenue to install increasing large backhaul links, which will mean the effect of the small number of heavy users can be mitigated. Another is to use systems to control the amount of bandwidth any single user can consume.
It has been suggested that that Tiscali are actually using software on its ATM network to try and shape the way bandwidth is allocated. The method used is that User connections are tagged into two priority groups, and depending on how much you use the connection you will move between the two groups. Users who hardly use the connection will have higher priority than the other users. It is thought that Tiscali set the minimum speed low priority users are allocated at 10kbps, with the remaining bandwidth then shared between the other users. If the high priority users are not using all the bandwidth available, the remaining bandwidth is thus shared between the low priority users. For those interested in what we think is the precise algorithm used this link provides a longer explanation. Of course if you are a 150kbps user, and end being given a low priority your speeds are potentially very low.
There are pitfalls to the algorithm, if the time between checking how to tag a user is too short, someone watching a 10 minute streaming video may find their video slows down to the point of being no use, and then have to wait a few hours to get their speed back to normal. The likelihood is that other providers are using similar schemes, but perhaps due to the way Tiscali advertises it has drawn in a slightly higher proportion of the 'all-you-can-eat' brigade, who feel it is their right to download 'the Internet' at 5GB per day for the next 10 years.
Bandwidth upgrades should mitigate some of the problems, particular on exchanges that perhaps currently just have a 1Mbps backhaul. In the past the churn of heavy users away from a service has been seen to help as well, but one persistent problem to date, is that Tiscali do not appear to be the most co-operative ISP with regards to migration and the ceasing of a service.