Network engineer Glen Turner has voiced his concerns about the Tassie filtering trial. In the comments section of my earlier post, Glen says a critical flaw exists in the trial’s testbed design:
’80-90% of Australia?s Internet traffic goes to the USA, a considerable distance and a large amount of latency. This is not simulated in the network design – rather the testbed assumes that all the content is local to Australia and thus of low latency. Effects like jitter and loss become exponentially worse as latency increases, so the testbed understates by orders of magnitude the performance effects on traffic. This single error is so huge that its effect swamps the results, rendering the performance results of no value at all.‘
Glen also says that the trial focused on the performance of the Ethernet protocol, when it is the performance of the TCP protocol which dominates a user?s experience when encountering a bottleneck like a filter.
‘TCP performance is determined by latency, jitter and loss. Nowhere does the report measure the effect of the insertion of a filter middlebox on those essential variables. In short, these amateurs have measured the things that don?t matter and not measured the things which do.’
Read more of Glen’s concerns in the comments at the bottom of this page.
