NNSquad - Network Neutrality Squad
[ NNSquad ] Re: Free Press: FCC used 'flawed data' in broadband plan
Hi George, When I mentioned limitations of speedtests, I think you misunderstood. They *overstate* the problem in that there are multiple reasons for low-speed tests (more than you mention, i.e., alternate downloads). These are mostly end-to-end tests, and there may be an issue on one or more hops BEYOND the ISP. We mainly want to test the ISP and their peering - end to end is many providers. I don't know enough to comment on the ofcom studies, so I will not. Let's go back to "data": what a p-value is. We (scientists) often use p < 0.05 for Significance. The catch is the non-linearity when we examine outliers, and the fact we are only testing a null hypothesis with a 5% chance of not being met. But it's an *average* number we are testing, and so a bunch of faster speeds than (say) 3 Mbps along with 5% unlikelies (who could be much lower) indicates that a measurable fraction of people don't get 3 Mbps. Disreali and then Twain said it right about statistics. Rahul p.s. I will restate my point about "anecdotes" - I have a single *data point* about my DSL speeds. It is not an anecdote - re. someone who says they have seen Elvis, there is a difference between that and my statement. On Tue, Apr 27, 2010 at 1:18 AM, George Ou <george_ou@lanarchitect.net> wrote: > What part of "show me the data" or some data or ANY data do you not > understand? > > I keep presenting my data and citing official studies like the one from > Ofcom in 2007. You (and the others in this thread) on the other hand is > making some fairly serious accusations without any evidence. I keep asking > for data and you guys talk about everything else but the data. Lauren > claimed that his cable provider provides less than half the speeds, and I > directly asked him for a speedtest.net test URL but I haven't heard a peep > back from him. > > Anecdotes are useless because I can find you a million anecdotes that Elvis > is alive, or that W. Bush blew up the tower, or that Obama wasn't born in > the US. The problem is that millions of these anecdotes don't mean a thing > until you show some actual concrete evidence. > > "I don't have data or a speedtest (not that those aren't without flaws, > including exaggerating the problem in some instances)." > > OK stop right there. What you're telling me is that you have nothing so > you're not really contributing anything to the debate. If anything, you're > contributing noise. > > Speedtest.net could be problematic but any problems would underrate the > performance of a broadband connection (e.g., I had another application in > use during the test), so it would err on the side of unfairly blaming the > ISP rather than making the ISP look better than they actually are. That's > why I prefer passive monitoring e.g., > http://www.digitalsociety.org/files/gou/usage-graph.png, and why I think the > FCC should use passive tests http://bit.ly/aSk96o. You also need continuous > ping tests to check jitter > http://www.digitalsociety.org/files/gou/East-coast-jitter.png. > > > "Scientific? Maybe not. Real - I would like to think so." > > I'm beginning to see the problem here. You view anecdotes as a substitute > for data and science. I disagree with you. > > > > George Ou > > -----Original Message----- > From: Rahul Tongia [mailto:tongia@cmu.edu] > Sent: Monday, April 26, 2010 12:24 PM > To: George Ou > Cc: nnsquad@nnsquad.org; Dan Gillmor; Larry Press > Subject: Re: [ NNSquad ] Re: Free Press: FCC used 'flawed data' in broadband > plan > > At what point do anecdotes not matter at all? To the person who > doesn't get what they (reasonably) expect, even a single "outlier" is > meaningful. Gee, 3 people had no problems, so thus I shouldn't > either? I switched back to the "regular" DSL at 1.5 Mbps (768 up) and > the VoIP worked fine. Scientific? Maybe not. Real - I would like to > think so. I don't have data or a speedtest (not that those aren't > without flaws, including exaggerating the problem in some instances). > > BTW, I am not picking on any one company (again, no names!). In some > ways, wireless is vastly more flawed for advertised speeds - but I > think people realize and even expect that. > > Rahul > > On Tue, Apr 27, 2010 at 12:43 AM, George Ou <george_ou@lanarchitect.net> > wrote: >> "is that the 128 kbps uplink DSL solution just didn't work well for the > VoIP >> service" >> >> And what I am saying is that absent any kind of data, you cannot make any >> generalizations about 128 Kbps upload service. I personally know 3 people >> with 768/128 Kbps service in California, and none of them have any of the >> issues you describe and everything works well and latency remains low. >> There are some known issues with some routers and Skype, but those can >> usually be addressed with software updates. But your generalizations > (again >> without evidence) aren't very useful at all. >> >> As for the delivered versus advertised speeds, I'm a bit tired of all the >> anecdotal evidence. I haven't seen a single person back up their claim > with >> at least a single speed test such as >> http://www.speedtest.net/result/794256737.png (my result at 13K feet with > 3 >> Mbps sync rate). As I've pointed out, the 2007 Ofcom data shows that US >> broadband providers deliver the closest throughput to advertised speeds > and >> this is consistent with all of my testing. >> >> To sum it up, please post data or stop making these claims. >> >> >> >> George Ou >> >> -----Original Message----- >> From: Rahul Tongia [mailto:tongia@cmu.edu] >> Sent: Monday, April 26, 2010 11:56 AM >> To: George Ou >> Cc: nnsquad@nnsquad.org; Dan Gillmor; Larry Press >> Subject: Re: [ NNSquad ] Re: Free Press: FCC used 'flawed data' in > broadband >> plan >> >> Hi George, >> >> I am not making any claims - "management" was just one possible >> (theoretical, and, esp. at that point in time, unlikely) explanation, >> but what does remain is the fact that the 64 kbps (which I tested to >> be truly 64 kbps) worked very well in India. Simple traceroutes from >> Delhi to global sites showed the system hops were very international >> gateway centric (which might be a good solution for some developing >> countries until they grow domestic/local content). >> >> The fact remains, and I am not attempting any explanations, just >> possibilities, is that the 128 kbps uplink DSL solution just didn't >> work well for the VoIP service. There's a similar analogy to the >> people who point out *they* cannot get the advertised speeds. The >> answer doesn't matter for the lay user - even if it's a good, >> reasonable, and technologically acceptable answer. >> >> Rahul >> >> >> >> On Tue, Apr 27, 2010 at 12:17 AM, George Ou <george_ou@lanarchitect.net> >> wrote: >>> Are you accusing your VoIP provider of sabotaging your VoIP service? If >> so, >>> please present some data and I'll be happy to publish the results as will >>> all the mainstream news sites. There are plenty of editors who would > love >>> to make something of it and plenty of public interest groups that would >> love >>> to take this information to the FCC. If you don't have data, we can do >>> without the insinuations. We can certainly do without the > generalizations >>> that Indian 64 Kbps DSL is better than American DSL because we can find >>> extremely bad outliers of DSL service in any nation and any city. >>> >>> >>> My mother had a 768/320 link (due to extreme distances and maybe some > line >>> quality issues) and she never had VoIP problems and we do video Skype all >>> the time. Even using uncompressed G.711 CODEC, you only need a little >> more >>> than 80 Kbps up/down (including packet overhead). >>> >>> What will absolutely cause problems is P2P traffic and to a much less >>> frequent extent bursty video streaming (like YouTube which sometimes >>> aggressively caches ahead). It might also be a case that you had such a >>> poor quality link that you were suffering extreme packet loss. I've > never >>> come across a DSL line so bad that VoIP would have problems and your >>> "evidence" sounds anecdotal at best. You don't even provide any >> continuous >>> ping results or even a few speedtest.net samples. >>> >>> So barring some unusual problems that you haven't documented in any way >>> shape or form, 3 Mbps broadband will comfortably run just about every >>> application on the Internet. Even 768 Kbps broadband will support 480P >> mode >>> Hulu (I measured around 500 Kbps). >>> >>> >>> >>> George Ou >>> >> >> >> > >