Are Openworld Port Throttling?
Friday 02 November 2001 10:12:00 by
After getting users of the BTopenworld service to monitor speeds for over 3 days, it appears that they are to some extent being throttled.
BTopenworld are denying any throttling or balancing the traffic at present. At our meeting with them they insisted that port throttling wouldn't be occurring. So what is happening? We know that
BTopenworld give priorities to different types of traffic and assign a route through their network for it -- The problem is when the traffic grows to fill that route, individual users will
see a slow down. This is very different to the 50:1 contention that users will experience over the ADSL segment of the network which affects all protocols equally. The results suggest it is
something internal to their network.
The tests showed that http traffic appears very variable but generally starts to decrease in speed from 2pm in the afternoon until around midnight, dipping to around 10KB/sec. This was almost
exactly mirrored by port 5000 - a relatively unused port. Peer to peer traffic on ports 1214 and 6699 follow an almost saw tooth form, with the throughput been around 4KB/sec at 10pm after steadily
decreasing from around 7am in the morning. The graphs can be viewed here: port 80
, port 1214
, port 6699
, port 5000
. We can supply the raw x,y data (excluding user information) to BT on request.
Some other ISPs (mainly Nildram) have hit capacity problems on their gateway into their network and are experimenting with traffic balancing to give UDP traffic a better chance -- The difference
here is that the users are being kept informed and Nildram are very open about what is happening. The results may look similar but once things are working, I suspect the difference will be
a lot larger again. This is because BTopenworld has been working on 155Mbps upgrades to their home gateways for some months, and UDP gaming traffic has been poor for longer, whilst for Nildram it
has only been an issue for a couple of weeks.
Some users have been able to achieve better results than others at the same time. Exchanges and PCs have been ruled out as possible causes by testing with friends' account details, proving that
the Home Gateways or the ISP's internal network is causing the problem. In the case of the Home Gateway it is possible for two users to be getting two different sets of performance as the system
divides them into two groups and one of these groups may be using up more bandwidth and therefore causing congestion before the gateway is 100% full. i.e. It's a lottery.