Speedtest is unreliable. Whilst RDP'd onto a server sitting in one of our (the ISP I work for's) suites which has 100Mb of uncontested internet connectivity, and large amounts of unused backbone I've seen figures vary wildly between 60Mbps + and 1Mbps. I just don't trust that site any more.
That said, I'm with Virgin at home and I'm finding the service to be highly changeable. Some evenings we're hitting 400k/sec download speeds, others I'm fluctuating between 2k/sec and 250k/sec even going from Virgin's own mirror server (ubuntu.virginmedia.com) Last night was a good night. Even though Feisty Ferret (7.04 of Ubuntu) was released yesterday and all mirrors were being hammered, I was getting 450k/sec downloads from the virginmedia server.
drak3 said:
In each country there are the central nodes which all peoples packets go through before they get out of the country so to speak. At the central nodes, there are underwater, land, sat links to the destination country's central nodes. Now IF the link from UK to France failed, then the routing protocols (algorithms to route packets) take over and determing another route to send your packets. However, since the new route might be through 3 other countries it takes longer. Usually the new route will cause overcrouding on the router (now the router serves its own people and the temprary) and most probably queuing will form (if router reaches 100% capacity). Its very complicated really, but the main point here is due to the compexity of networks nowadays a small failure propagates so fast and causes a lot of problems. The more money the ISP spends on backup solutions the less problems. However, renting twice the capacity on a fast optic fibre link "just in case" is not something any ISP will do for home users at least. Instead for home users, will have an inferior cheap backup link to compensate for the x hours downtime on the main.
ps: This scenario might happen on any side of the connection and have the same side effect. Your ISP, or the server's.
Interesting view of things, but not entirely accurate.
Most ISPs inside the UK are part of the LINX exchange, including Virgin (NTLI ASN:5089). Accessing pretty much anything inside the UK
should go via the LINX exchange, inlcuding SpeedTest's servers, unless they've picked one of the most minor hosting firms they could find. Every ISP will have
at least one link to either Sonia, Telia or Abovenet; the three main suppliers of internet backbone. Speaking from experience, losing a link like Telia has resulted in barely even a 10ms increase in lag as the network very, very quickly adapts to the change.
For emphasis: There is no such thing as "the link between UK and France" You're talking about lots (and lots) of redundant links. If the link Sonia provide goes down, you can use one provided by Abovenet. Each of the major backbone providers are part of the LINX consortium too so if your particular link to them died you could go through the LINX.
What is more likely to increase latency is one piece of equipment going very slightly wrong and causing the lag itself. The route is still valid so the routing doesn't get updated at all to avoid it unless the network admin picks up on the issue and forces it themselves. One of Telia's interfaces started doing that to us about 4 months back, just for accessing a specific data centre in Italy. One quick message off to the network admin and all was fixed from our perspective within a couple of hours.
In this particular guys case I'd be more inclined to believe that speedtest's server was being slow or being hammered than it bieng due to any links going down.
ouch
