MWEB's Traffic "Shaping" Visualised
Like most people here that are religiously following this thread in the hope of getting some answers, I have spent the last few days running tests of my own at home to try and get a better understanding of what is happening to my MWEB bandwidth. One of these tests is to graph my traffic so that I can get a visual picture of what's going on over time. Something I had noticed previously was that I was being limited per connection as opposed to my total throughput, and this meant that my first attempt at graphing was skewed when my download manager, by default, used multiple threads. For example, I had noticed that during the day (and this applied to every international URL I tried), I would be getting about 256kbps (30KB/sec) per thread. If I used 4 threads, I would get 4 x 256kbps total throughput. I wanted to show this phenomenon over time on a graph, so I decided to create a large file on my webserver in Europe and download it using a single threaded HTTP connection overnight... The results can be seen below, so lets take a look and see what conclusions we can draw from it.
* 17h00: I begin the download. In realtime on the download manager, I'm seeing fluctuation between 20KB/sec and 40KB/sec, but the 5-minute average you see here shows the true speed of about 30KB/sec (256kbps).
* 22h00: Here we start to see the first signs of change as the download gains momentum, peaking at just below 1mbps at midnight, but then strangely dropping off to 512kbps again.
* 01h00: What have we here?! Suddenly the taps are open and I'm getting full speed! I say "full speed" because remember, this is only a single connection and even in a perfect world, one generally doesn't see a single connection to an international server achieve > 3mbps (unless it has been transparently cached locally of course).
* 08h00: Bandwidth throttling comes down like a hammer and within an hour I am limited to 256kbps once again.
Conclusions:
1.) What this graph shows beyond a shadow of doubt is that I am subject to hard-throttling per connection. This is *not* mere traffic-shaping, rather these are hard-and-fast rules that have been put in place at some point along the way.
2.) If the Technical Manager at MWEB doesn't know about these limitations that are obviously active, then they have a serious problem with internal communication. However, it is far more likely that he DOES know about it, but for whatever reason, he cannot come right out and tell us about it. This is probably the most frustarting part about all of this... It should not fall to customers like myself to have to do investigations like this in order for us to get the full picture. If we are being throttled based on time of day, then we have a right to know about it, and clearly, *SOMEONE* at MWEB has put in this limit.
My Theory:
Something I haven't mentioned yet is that I have two friends with MWEB accounts in the Durban area who are experiencing the same problems. However, and this is especially perplexing, two other friends in Cape Town are having NO SUCH ISSUES. At any time in the day, they are seeing single connection speeds of > 200KB/sec. Its worth noting that they have similar usage on their accounts too, so it would appear that the throttling isn't based on bandwidth usage, but rather geographical location and / or the route that the traffic takes to get to its destination. Quite likely, my Cape Town friends' traffic is travelling by different backbone provider compared to mine, and therefore subject to different bandwidth management. Again though, MWEB should know this; we shouldn't have to tell them.
When I find some time to log a fault with Telkom, I will submit the support document that has been posted here recently. It will of course prove less useful than the information I have provided above, because as we all know by now, local traffic is 100% so MWEB speed tests will reveal nothing, and simply dialling up with an alternative ISP validates that the line is not to blame.
Either way, I hope that this information proves useful. My next test will be to dial up with one of my Cape Town friends' accounts to rule out (or confirm) the theory that it is only a block of accounts that is affected.
10+ isp's in one year isnt fun
