This is from Netflix, it downloads Netflix content and reports the speed back.
This is important because unlike your average Internet speed test (which ISPs take pains to optimize), there's a very real possibility that your ISP is happy to let your Netflix experience suffer - assuming they don't throttle it outright - as previously mentioned on HN:
I've had a tech who came to deal with a reported issue I was having tell me to test on a specific Time Warner Cable speed test. He then got upset/flustered when I used a different speed test to demonstrate that the issue was still present. I had to explain that I use the internet connection I was paying for to do things on domains that aren't owned and optimized by TWC.
To give the tech the benefit of the doubt, it's possible that he just isn't certain he can trust other speed tests. They may point to issues beyond his control.
Absolutely. The tech has a limited scope and typically doesn't have the ability to escalate engineering issues. You need to keep complaining and Twittering until you get a senior guy.
In my case, the TWC guys replaced every piece of coax and splitter from my cable modem and TVs to the pole.
Turned out the issue was the changed out the cabling on the big avenue near my house and used the wrong grade of cable.
I called the TWC state lobbyist and told him I represented 150 homeowners who were filing a complaint with the regulator and the city council just head of their franchise agreement renewal. Trucks were replacing those lines the next morning.
> In my case, the TWC guys replaced every piece of coax and splitter from my cable modem and TVs to the pole.
I am surprised by the coax inside the house. Usually they'll help if they can, but your house wiring is your problem. (Unless their solution is stapling a coax cable to your baseboard.)
The techs were observing erratic behavior, but there wasn't any evidence of bad cabling -- I had actually replaced it about 3 years before. Also, my wiring is low-complexity... I terminate everything in the basement, and have 1 TV. The TV cable ends about 6 feet from the cable box.
Even in other circumstances, the techs are pretty good.
Yeah, it's tough to find the line of when to stop giving the benefit of the doubt though. I was using speedtest.net at the time, which is pretty well known and widely used. I had also had multiple visits recently regarding the issue that still had not been resolved, and so my impatience made the decision for me.
I couldn't find any non-speedtest sites where my latency was less than 100ms, but Comcast's own speed test, several other public speed tests, speedtest.net, and even ookla.com (the company behind speedtest.net) were perfect (the ~20ms you see on the graph).
Speedtest.net and other speedtests often choose the speedtest server that's close to you and has low ping.
What you describe is probably congestion in the network somewhere that isn't the last mile. Speed tests aren't a good indicator for any issues occurring anywhere but the last mile, as they won't take the same path with fairly high probability.
Given that he has a nice graph with hundreds of data points, I'm guessing he got the ping results from the ping command, not by running the speedtest.net test(which would actually select the lowest ping server)
You're right, though I had assumed he was pinging whatever server speedtest had selected, which may be erronous.
Really, my point was that congestion is often localized in a network. The fact that it isn't present in a path doesn't mean it's an ISP gaming speed tests - though perhaps they'd be more keen on peering directly and avoiding congested transit if it makes them look better, and I would absolutely believe that would happen.
I think the argument is that the speed tests are testing ideal/non real-world usage. That seems an awful lot like gaming the results. A more realistic, indicative speed test might hit several servers at points around the country, or even world. Most of my internet traffic is not hitting a server on the outskirts of town or in the next small city over.
Ugh. That's terrible. I kinda figured that it's an eventual conclusion though, so when speedtest.net starts giving me contradictory results to what I am experiencing I have a pretty good idea of what the check first.
On a separate note, what tool did you use to create that graph?
Looks like just redirecting a recurring ping command output to a text file then graphing the results in google docs or any other spreadsheet. I've done the same to shame Comcast.
Ooh, a chance to plug a toy side-project! I wrote "pingd" [1] to ping an address continuously and dump (timestamp, latency, dropped) tuples to a SQLite database, and also report results (tables only, no plots) via a web interface. Slightly less hacky than "ping host > logfile". Also to shame Comcast.
Just today I switched back to TW from Frontier(FIOS). Using fast gave me more than double the speeds of fios. The tech was super cool and he normally uses speedtest but I told 'em, try fast. Plus now I finally get to use my brandnew surfboard modem and brandnew RT-N66U router.
I have TWC in Brooklyn and for a while was on the phone with Level 3 support often. Every time they had me use speedtest.net. So even their network engineers trust that site more than their in-house one.
I agree to a point. If you use speedtest.net and show several different testing sites and the results are consistent, that should be good evidence. And you need to use a hard-wired connection because a lot of peoples wifi isn't very good.
>Former telephone technician(* ) here. What I'm interested in is the connection from your home to our PoP.
That's only relevant if the complaint is about the connection from your home to the PoP. If the technician was sent out to solve an issue with the Internet being slow in general, he/she needs to actually find what the bottleneck is, not call it a day when some traffic makes it to a PoP.
It's a stupid idea to do a truck roll to resolve upstream peering/transit/congestion issues with the IP network. The technician's job is to diagnose OSI layer 1/2 issues to your local DSLAM or CMTS.
With Time Warner, there is an absolutely impenetrable firewall between customer service and backend/upstream network engineering. Rolling a truck appears to be the absolute ceiling on what a call center employee can do.
This is unfortunate, but I can sort of understand why. Customers are frequently mistaken on what's wrong with their connection, example, the customer thinks they're seeing 0.5% packet loss and high jitter consistently to something that's 3 hops upstream of their default gateway, when in fact there's something messed up in their home Wifi which is on channel 1 the same as their four neighbors.
It takes a lot of time (time = money) to train first tier customer service reps how to recognize what needs to be escalated to the people who have 'enable' on the network equipment 2 to 4 hops upstream of a typical cablemodem/DSL customer. And then the management decision to empower them to do so if necessary. And a method for the NOC of filtering through the crap to discover that "Yes, this customer really has discovered some service impacting issue that has not been automatically reported by our NMS".
Usually because a customer can't properly articulate on the phone what's wrong with their connection. Last mile broadband providers want to do as much as possible to avoid a site visit to a home. Unless they're charging $95 per visit one technician visit for 30 minutes can eat up all the profit from that customer for an entire year.
How much does your ISP charge you? Verizon managed to get $71.48 a month out of me on a connection that had been set up by a technician nearly 10 years prior before I got fed up. A technician visit at that rate would probably eat just 1 month of profit on a connection whose initial installation cost had been paid many times over. The reason that I got fed up was that they kept trying to get more. Nearly all of those years had involved a triple play, for which they had raised rates until they charged ~$162 a month.
Sadly, I was not the one who negotiated with them. Had I been, I would have been fed up well before things became that bad. The $71.48 came after breaking the bundle I favor of an OTA antenna and VoIP. I managed to talk my way out of the ETF by telling a supervisor that the ETF was the same whether I cancelled all services or just 2 and that I was giving Verizon the chance to keep some business, provided that they reduced services to just the Internet connection and waived the ETF. The supervisor waived it.
After breaking the bundle, I took over responsibility for such bills. I am paying Cablevision ~$45 a month now. Bandwidth is lower, but I reason that as long as bandwidth is high enough to deliver traffic with acceptable latencies, being able to burst higher does not matter that much. So far, things have been okay on both pricing and having a usable connection, despite a few initial goofs by Cablevision.
How much profit do you think Verizon actually makes from your $70/month connection?
FiOS revenues make up 70% of Verizon wireline's revenue, and the average monthly bill is like $110. But the division's profit margin is 2-5% most years. Let's say generously they're making $7 a month on you (10%). A Verizon tech makes $35/hour+ not including benefits. So a one-hour field call wipes out 5 months of profit.
They make nothing off it because I cancelled service.
Google searches show that they have had a special $29.99 reactivation rate for years. Several months ago they made it $39.99 in the second year. A couple years prior, they had also added a $1.99 municipal construction fee. That works out to be $37 a month. I assume that the amortized cost of keeping the fiber lit is below that because I doubt that they would make such an offer at a loss.
Charging a long time customer higher pricing would provide them with money that they can spend on things like Fios network expansion and network upgrades, which are expensive.
Your analysis is wrong. When Verizon spends a bunch of money to wire up a neighborhood, that doesn't come out of profits. Instead, it books a capital asset that is depreciated over time. Verizon wireline's EBITDA margin these days--five or six years after the end of significant network expansion--is about 22%. That's what's left after paying for ongoing maintenance, support, and bandwidth, but before accounting for recouping the cost of the original build. After accounting for depreciation, interest, and taxes, it's under 5%. That means the up-front cost of your connection is being recouped at $10-20 per month at most.
Verizon spent about $20 billion on FiOS, and has about 7-8 million customers. So the average customer cost about $2,500 to wire-up. That means it'll take 10-15 years to recoup the initial cost of the build-out.
The special-offer pricing is not meaningful. Most of the cost of building and maintaining fiber is getting to the neighborhood. FIOS penetration is only about 1/3 in the places where service is offered. Which means that Verizon has a lot of incentive to make offers to those 2/3 at a rate below what it could sustain charging everyone.
Again, even though the average FIOS subscriber pays $100+ per month (triple play, etc), the division's net profit margin is under 5%. It would be losing enormous amounts of money if people were paying $30-40/month.
Your analysis is wrong because the subscriber count excludes Fios subscribers in areas that they sold to Frontier while the $20 billion includes them. The costs also vary depending on subscriber density. Furthermore, the actual installation cost is $750 in my state according to Verizon:
I recall that number was ~$1000 9 years ago. I find it hard to believe that Verizon did not recoup that over 9 years. That is less than $10 a month over that time frame. My household paid $10 to $20 a month more for higher speed tiers over that 9 years. Those cost Verizon almost nothing so it is reasonable to conclude that they recouped the initial installation cost in this instance.
Also, Verizon has two tiers of special rates. One for new sign ups at locations that have already been wired and one for new sign ups at locations that have not. Mine is one that has been wired. Entering into an arrangement where the income from providing service is less than the day to day expenses of providing it before including one time installation costs even with a contract period makes no sense, so I do not think that Verizon is doing that. Since the one time installation costs are paid, they would have been making money even at the lower rate in my case under that assumption. Obviously, it would not be as much, but I am okay with that.
I realize that individuals' calculations often underestimate ISP expenses. However, my case is one where such calculations should apply. I also realize that wireline margins are low in general. I did not mean to state that the profitability of my particular Fios connection was the norm. I only meant to say that there are instances of old lines where the margins far exceed the average. That is why I said the connection was setup "nearly 10 years prior". The profitability of the connection would have been a very different matter otherwise.
I took another look at this. The $750 per house figure was actually for wiring the neighborhood without going into any homes, which costs an additional $600:
>In New York, Verizon said it costs about $750 per house to bring FiOS service to a neighborhood, plus an additional $600 to run the fiber into the home of a consumer who wants to purchase FiOS service.
Assuming 40% penetration (also in the article) and $600 on top, the cost is $2250, which is close to your $2500 calculation. I still think Verizon made that up over the course of 9 years. I just need to include the set top box rentals in the calculation. Assuming that my household paid $15 on average over 9 years more than the base tier price on average for the internet connection and each of the 3x $5 a month set top boxes that we had for 8 years cost Verizon $250 each, I get $2310. This is before considering that those should still be worth something and the principle that Verizon would not offer lower rates than their operating costs. That would be a great way to obtain negative margins.
sadly, I am... I work for an ISP that generally doesn't make these mistakes, our network engineering/operations/construction team enjoys observing all the stupid things Centurylink and Comcast do in our local market.
OTOH - the technician doesn't need to be at your house if you have a good connection to the PoP. Field technicians may not be trained in troubleshooting network beyond the PoP and likely don't have the right access to tools to make the changes that need to be made.
I work for the CDN Fastly and we have something very similar to Fast.com.
fastly-debug.com.
The goal of our site is to gather debug information but I use it as a reliable speed-test to a good portion of the internet's websites. I think this is a good tool like fast.com because unlike the 'official' speedtest sites, if you optimize ours or Fast's you're actually making the internet much faster for a lot of sites for users.
Thanks for the tip. In this case the JS is performing a very important function, helping us to measure distances to multiple pops, so it is better to stay in. I will forward your point about the domain, but I think there is a reason for that particular domain.
Verizon FIOS in my area was throttling down Youtube and Netflix to unusable speeds. I was paying for a higher end connection speed and getting constant speed issues.
So then switched to a cable provider, pay half what was paying with Verizon, get lower advertised speeds yet have a much more reliable and consistent connection.
Also Verizon had free installation at that time when I signed up, so they installed a bunch of their equipment on premises, even a phone (including a battery pack) and now after I dropped them they come to door, in person, every other month begging me to reconsider.
Verizon did not throttle speeds. Instead, Verizon wanted to be paid for peering link upgrades. Such links are only free when two backbone providers feel that they both benefit from making them free. The division of Verizon that manages backbone is different from the one that runs Verizon Fios, so the fact that the other division had already been paid should not matter very much in that mindset. It was the refusal of both sides to budge and the growth of Netflix traffic that caused bandwidth requirements to exceed the network capacity. That is what people incorrectly called throttling. Setting up the Hurricane Electric tunnel broker so that Netflix traffic would be carried on IPv6 was a reliable workaround for me in New York.
Netflix might have been able to get Verizon to be more reasonable if they modified their streaming clients to upload random numbers back to them in excess of the amount downloaded. Then Verizon's logic would have dictated that Verizon owed Netflix's backbone provider money and Verizon would have wanted to avoid paying rather than be paid. Such a situation should have caused the links that Netflix traffic used to be upgraded without much negotiation. I was surprised that Netflix did not do that as a workaround. That would have made anyone even superficially familiar with game theory smile.
However they structured the deal, as an end user, it seemed like throttling to me. I could download ISOs and other large binary files ok. But somehow streaming from a few sites seemed broken.
That is one thing to expect that on a low end 2Mbps connection, but I was paying for a 100Mbps wouldn't expect to have any issues at rate I was having them. How they decided to structure their payments downstream is their problem so to speak (putting on my non-technical consumer hat on).
I am glad I had a choice to switch and now with half the price or so (and also half the advertised bandwidth (65Mbps) I seem to get a much better user experience.
I fully expect to be down voted for saying this, but the way that the Internet works means that you have no guarantee that communication between two computers via Internet Protocol packets will operate at either's last mile connection speed or even operate at all. When you get an Internet connection from an ISP, what you are getting is a connection to their network. Access to the other networks that make up the Internet is something that they both do not guarantee and cannot guarantee. The only reason that you get it at all is that if most of the others were not typically reachable from them, they would lose business to an ISP from which most of the others are typically reachable. If the links degrade, it only matters to them when they perceive inaction as impacting their profits more than doing something. In some cases, the ISPs cannot reasonably do anything about because the owner of the other network refuses to provide sane terms for doing something or even allow access at all. For example, China and the US were connected by a 155Mbps link 10 years ago because Chinese ISPs wanted non-Chinese ISPs to pay for transit connections at >10 times the going rate. In the case of the 25.0.0.0/8 block, the owner is the British government, which refuses to allow interconnection. Consequently, it is simply not possible for any commercial ISP to provide quality access to every other network on the Internet or even provide access at all, even before we get to the cases that cerrain ISPs are jerks, like Verizon was with Netflix.
I was affected by that peering dispute too. That lead me to research it, understand it and workaround it. Consequently, I use that knowledge to implement workaround whenever a peering issue affects me.
To name one, I observed 1.3 second average RTTs between Shanghai and New York when I was visiting family there in 2014. My research into the Verizon peering disputes prepare me fairly well for dealing with it. I initially used OpenVPN over TCP and my pings were running inside the VPN tunnel. Packet loss was typically in the range of 10% to 20%. There were occasional latency spikes to 30 seconds. Switching to UDP lowered latencies to ~300 ms in either case, but bandwidth was rather low. I forget how bad it was in absolute terms, but I can say that it was too bad to use YouTube or even VoIP. What helped was routing through servers in Japan.
The trick was to do a perverse variation on a NAT where the LAN and WLAN interfaces are the same and the NAT is effectively a proxy for a port on a remote system. It works well for routing around peering disputes when you can get a VM in a place where there are not peering disputes between the local ISP and the VM provider and the VM provider and the destination. Trace route results at the time suggested that Japanese ISPs appeared to maintain both sides of the underwater link between them and China while US ISPs only control the network sockets at the Internet exchanges in the US. Routing through the Japanese internet backbone presumably eliminated the congested US-Chinese peering links from the path, which showed because packet loss went into the single digits. I do not remember the exact numbers, but it also put speed between 1Mbps and 10Mbps. That was enough for me to make VoIP calls to NY while I was visiting family in Shanghai and YouTube was usable and even load YouTube videos.
However, the connection was still bad enough to make Verizon's peering dispute with Netflix look mild in comparison. There were occasional latency spikes to 30 seconds, which appeared to be from buffer bloat in the great firewall. Google search results on the Chinese backbone at the time suggested that the entire county also appeared to have a star-like network topology for its backbone, with the exception being that Hong Kong was on the outside of that. My trace route results from Shanghai then were consistent with it.
Another OSS developer in Shanghai informed me at the time that the three major ISPs in China have peering disputes among themselves despite the Chinese government owning all three. He claimed that communications over his 100Mbps last mile connection at his house with his employer's office operated at dial up speeds because of that. I did not confirm that, but I believed him. What I saw of Chinese Internet connectivity while I was ther struck me as being incredibly terrible.
I still remember how bad things were with Verizon and Netflix. However, the reality is that things could have been many times worse like they were in China when I was there in 2014. At least with Verizon and Netflix, you had the option of the HE tunnel broker. In China, I saw ISPs using carrier grade NAT, such that using IPv6 via the tunnel broker is impossible. That is not to say that IPv6 via the tunnel broker is some sort of magic cure for issues. The only reason IPv6 via the tunnel worked during that dispute was because there was no congestion between Hurricane Electric and Verizon as well as between Hurricane Electric and Netflix. If many people had tried using it, it would likely have degraded too.
Whether it's what they intended to sell or not, ISPs, through their own marketing material, sell a connection to "the Internet", and unfortunately for them that term cannot fairly be applied to just Comcast's network anymore (though it would make for a comical court case).
If they can't provide what is advertised, specifically "Internet access", then it's false advertising and I expect, and will fight for, my money back.
The real issue is that those of us that are not ASN operators are at the mercy of those who are. The only true solution is finding an ASN operator that has user interests at heart. Municipal fiber is probably a good example of that. There are other ways (all involving getting an Internet backbone connection), but they are not cost competitive with last mile distribution.
A workaround would be to rely on an IPv6 tunnel broker or a VPN tunnel to a VM in some datacenter. As long as the tunnel is unaffected by peering disputes, your traffic over it will live at the other side of the divide between content producers and consumers. That gives a far better experience than relying on a last mile operator to give you access to the broader Internet.
Speaking of which, does anyone know when hacker news will support IPv6?
I am paying to connect to the Internet, if it's slow Verizon fucked up end of story. If Verizon needs to pay netfix extra to get done then fine. Instead Verizon assumes they have full monopoly power and does not care about service, which suggests they need to be broken up.
Such problems are enabled by the mix of having multiple organizations responsible for the path, there being more traffic than it can handle and the organizations not coming to terms on how to handle that. In the case of Verizon and Netflix, the division between Verizon's enterprise and residential services meant that the enterprise division thought that they had not been paid for the traffic.
How would breaking up Verizon solve a problem caused by Verizon's internal divisions being separate?
The garbage return traffic scheme would only work if Verizon and every other ISP were completely asleep at the wheel. The public has enough trouble understanding the network neutrality debate without reading the headline that Netflix is intentionally wasting petabytes of backbone bandwidth every day.
The idea of ingress and egress being roughly equal as suggesting that upgrades should be done at no charge by either party to the other is how peering agreements often work. If the traffic levels are equal (or even skewed in the other direction), on what basis would they ask for money?
Well of course they do, the number of people for whom heavy LTE usage is an economically feasible prospect (i.e. unlimited usage) is vanishingly small. Since their network can't tell who pays per GB and who doesn't, yes they want people to consume as much data as possible.
Verizon FIOS was doing this about a year ago in the NYC area. I was getting crappy YouTube and Netflix playback but everything else was fast. My solution was to use a VPN service, and suddenly I was magically streaming Netflix and YouTube in crystal clear HD.
I think what was happening is that the VPN service had servers in the NYC area that had great connections to the local ISP network, and they also had a great connection to the Netflix servers, vs Verizon was deliberately keeping their pipe to the Netflix servers small so that there was limited Netflix bandwidth entering the local ISP network.
Setup the hurricane electric IPv6 tunnel broker and you should be immune to disputes between Verizon and Netflix or YouTube as long as the peering links between Hurricane Electric and Verizon remain congestion free. Both Netflix and YouTube traffic will run over IPv6 and HE is a tier one ISP known for aggressively peering with everyone. They would likely be the last ISP to try to charge Netflix and YouTube for connectivity.
They're not actually begging you to come back. They just pay someone shit wages and commission to engage in real-life spam. If they really gave a shit, they would not have lost you as a customer in the first place.
Why improve business fundamentals when it's cheaper to manipulate human emotion through marketing?
Well they paid enough someone to have them drive out to my house. I can see how in their eyes they lost a deal because house already has their equipment installed and I only was a client for 2 years. But well it is their fault.
I guess my point is, how does a company prevent its customers from turning into statistics on a report? You were part of the churn on some report somewhere in their giant number crunching machine.
It's cheap to have someone pester you and 100+ other people that day to see if you'll budge, but it's far from a personal effort to win you back.
For what it's worth, I've had Comcast nearly 4 years and I'd say 80% of the time, at least, the speed is great and I get more than what I pay for.
Occasionally at really odd times (mostly middle of the night) I'll get a disconnection from service, but that hasn't happened in a long time, and maybe twice in the past 4 months my speed has dropped pretty sharply. But for the latter, when compared to the service I've gotten the entire time, it's just a blip.
The customer service is absolutely awful though, unless by some odd luck I can get forwarded to their corporate offices in the US.
I work from home, which means I'm hyper-sensitive to my internet quality. I also sometimes work through the night to early in the morning.
I've found that my Comcast connection here in San Francisco fairly reliably drops for a few minutes every day at around 2-3 am. Then it comes back and works find again. I've thought about calling them about this before, but I'm not sure that I want to spend a couple hours on the phone convincing them it's a real issue.
I had that happen on my Comcast connection but it was always between 2-3pm. It wouldn't drop completely though it would just suddenly start dropping 95% of the packets and ping times would go up to 2-3 seconds. I pestered them about it for a month and eventually a tech happened to arrive while it was happening, he called their CO and they found they had a network switch that was intermittently malfunctioning. After that it never happened again and for helping them troubleshoot it they gave me 4 years of free HBO and the cell number for one of their engineers so I wouldn't have to sit through tier 1 and 2 support again.
I live near Atlanta and I get the same issue around the same time. It's not reliably every night (thankfully, since I'm on almost every night) but it is at least once or twice a week, sometimes more, for 15-45 minutes. I can see in the modem logs where I'm losing my connection to them and the house phone also gets disconnected.
I'm in the same boat as you where making the phone call feels like a huge hassle. But it's interesting seeing a few other people with this same problem, maybe I should call them after all.
That drop is what I was describing in my other comment. Happens generally between 1am-3am.
When I've called about it they say they can't send a tech out at night to test, so nothing has ever officially be done about it.
Sounds like more of a systematic issue though if you're seeing the same thing in SF (I'm in San Jose). But like I said, it's probably been a couple months at least since the last time I had that issue, or noticed it.
They are, but DNS is not used to redirect slow.com to fast.com. They both resolve to different IP addresses. It's only when the browser requests "http://slow.com" from the slow.com server that it is told to redirect to https://fast.com/. This is an HTTP redirect, and has nothing to do with DNS.
The context was "you may change it", where "it" is the redirect from slow.com to something other than fast.com. But he wouldn't be changing any DNS entry as he's still resolving the hostname to the same IP address; instead he changes the http redirect to some other site. No DNS entries need to be modified as part of "you may change it".
I'm not sure why, on a site called Hacker News, that you're surprised on being called out for a technically imprecise statement. Instead of digging in your heels with bogus justifications that you're rightfully being downvoted into oblivion over, why not thank those who replied and offered a more technically accurate explanation?
i'm not being downvoted to oblivion. if i were, i would have deleted my comment for fear of getting hellbanned. what happened is you just made that up, in your head, and assumed it to be true. just like the words you put into my mouth earlier.
LET ME ASK YOU THIS: even if it were refering to an http redirect, who REALLY has control of where slow.com redirects to? the guy who owns the domain. he can point it to any server or cname he wants, and then, maybe redirect browsers somewhere else, or just serve whatever content he wants. it could change day to fucking day, or hour to hour.
_because that's how dns works_
this entire thread is ridiculous. this is called 'pedantry', is usually misguided, and it very rarely leads anywhere productive. but you knew that, since you seem to be an expert in it.
judging from the number of positive points on my original comment plenty of people understood me. you're in the vocal minority who has trouble interpreting linguistic context.
For those hatin on speedtest.net and wanting upload, http://speedtest.dslreports.com/ and https://speedof.me/ have booth been around for a while. The reason for fast.com is that it tests download speed from netflix. ISPs can't prioritize it without prioritizing netflix as well.
>ISPs can't prioritize it without prioritizing netflix as well
There are all kinds of tricks they could use to game this. If they applied their throttles over a 5 min window for instance. I'm really curious what Netflix is doing to guard against that, if anything.
That can also be attributed to deep buffering. Your local router and even the next hop will make it look like data is being transferred instantaneously until their buffers fill up.
Download burst on the order of MB/s in the first few seconds can't be attributed to deep buffering. The router doesn't magically have a buffer of things you might visit.
I tend to use testmy.net, which includes both upload and download.
> The reason for fast.com is that it tests download speed from netflix. ISPs can't prioritize it without prioritizing netflix as well.
Is there any way to distinguish traffic between the two? Does fast.com actually download a random video from Netflix? Does it benefit from the Netflix local caching server? If ISPs can find a way to cheat, some might.
It downloads over https, so anyone watching the traffic can't really get much more than the domain. The download for me comes from ipv6_1-cxl0-c144.1.sea001.ix.nflxvideo.net which is, i assume, one of the Netflix servers in Seattle over IPv6
Netflix should advertise the speeds people are getting from them. And alternative ISPs if things are regularly slow.
Here in the UK we have fibre through one provider and lots through the phone lines because BT were forced to open their network. I get the feeling there is little choice in the US?
What I'd really love to see is this concept provided as a service by all of the big streaming/gaming/large-content-blob providers and aggregated into a single page.
I have absolutely no reason to believe that every well-known "speed test" app/site/utility out there isn't being gamed by my ISP. A speed test that showed me my actual streaming bandwidth from Netflix, actual download speed of an XX MB file from Steam, actual upload bandwidth to some photo-sharing service, and actual latency to XBox Live or some well-trafficked gaming service would be awesome.
There is good reasons for this, it can help a ton in figuring out where exactly the issue is. If the speedtest server is off-net, you have no idea where the problem lies. Transit provider? Peering? End network? my wifi? Last mile?
I'm not sure this would need to be managed by any of those companies though, I think everything you listed could be tested in a browser by a 3rd party. May require authorization to those services though.
Seems like this is really about training their consumers to define the quality of their internet by their reachability to the Netflix CDN nodes.. Smart move on Netflix's part.
I'm a software engineer with all sorts of shit hooked up to my 100mbps (thanks Comcast for no gigabit in your worldwide headquarters city of Philadelphia) downlink (still 10mbps up because you know...) and Netflix is absolutely my #1 concern at home.
Eh, netflix maxes out content at a fraction your comcast speed. Doesn't really matter if Netflix is going 50mbps or 100mpbs. Making netflix work is important, but I'd be pretty disappointed if my ISP was giving me 20mpbs, even though Netflix would never need that much.
And, having your pipe overloaded by things other than Netflix. I have two kids watching Youtube on pads and a wife watching TV in another room. With my fake Uverse (DSL) the TV uses the internet connection so, we are often streaming four different videos at the same time.
When I used to be a Netflix customer it was more the variability of my connection that was an issue and not its "speed" at a given optimal time.
Usually I could begin a stream without problems. But often while streaming (often enough for me to realize streaming was a bad experience) the bitrate dynamically dropped way down to a terrible quality in response to what I imagine were poor network conditions. Netflix no doubt sees this dynamic quality adjustment as a feature, and preferable to buffering, but I chose an HD stream and I'd rather even see an SD quality video that I could be sure would stay that quality than switching between HD and very low bitrate, fuzzy, artifacty video.
I don't blame Netflix for the quality of my connection, but streaming is just not as reliable as cable and it's not one of those Moore's Law type things where throwing more processing power or memory fixes the network issues.
Hi, I work on the streaming algorithms at Netflix.
You are correct that throughput variability matters a lot as well. The trouble is that the variability is somewhat unpredictable (at least, with our methods so far).
And you're also right that we view the "dynamic quality adjustment" as a feature! Without it, we wouldn't be able to serve both 0.5Mbps connections and 50Mbps connections seamlessly. But we don't like those sudden quality drops, either. We track how often that occurs, as well as how often rebuffers occur, and we work to eliminate both. We use large scale A/B testing to measure improvements in the field. If it's been a while since you tried it, I humbly suggest you consider trying it again. We've made a lot of progress.
If anyone is interested in working on this problem, we're currently hiring engineers on the Streaming Algorithms team with a focus on mobile. https://jobs.netflix.com/jobs/860465
Since I have you here, can I ask a random question? My kids love Puffin Rock, and it's the only show that I would have issues streaming. It would start out fine, but a few minutes into it start buffering and would take minutes; resume for a bit, then do the same thing.
When this happened I could switch to any other Netflix show or movie and it worked perfectly fine. When the kids whined, I'd switch back to Puffin Rock and the problem would persist.
So my question is: is it possible that either some shows are encoded differently so my player had trouble with it? Or is it that specific content was being messed with by my ISP?
The players in question were Roku 2 and 3, and my ISP is Charter (business plan). I haven't tried this in a few months since it got really annoying, but can try again if it helps.
Another Netflix engineer here - have you tried with something not popular? It may be that the other shows you checked were cached via OpenConnect but Puffin Rock wasn't.
My email is my username at netflix.com. If you can send me the email address connected to your account, I can see that your problem gets to the right people. (Disclaimer: I'm not on the OC or related teams directly)
I also have a question. According to fast.com, my connection is 270 mbp/s (which jives with the official 500 mbps I'm getting). Yet, I've noticed that on chrome on mac, when I launch a new show, it often starts at a very low bitrate and then gradually switch to a higher bitrate. The problem is that this happen on every episodes.
One workaround is using the ctrl-option-shift S combination and manually set the bitrate to the highest settings but it's still a bit annoying.
Have you tried the "Adjust your Netflix playback settings" advice at https://help.netflix.com/en/node/11559 ? Not that I haven't seen this adaptive stream behaviour, but I find it affects my mobile devices (like iPads) far more than on a wired, desktop computer.
You could also try it on a PC/laptop and then check other streaming servers (right CDN column) by pressing Ctrl-Alt-Shift-s. Or check the current stream quality / buffer state with Ctrl-Alt-Shift-d.
I'm frequently on a network that performs below 2.0 Mbps (with fast.com recording it as 2.0 Mbps, 1.6 Mbps, and 900 Kbps), generally with around 20% packet loss on the third hop (Verizon, now Frontier).
Netflix streams perfectly. It is astounding. No one else comes close. Amazon Video and YouTube work around half of the time, with rebuffering; video from anyone else—Vimeo, Apple, and any network's player included—is basically non-functional.
So, thank you all so much for excelling at what you do! It's very appreciated!
I'm working with some ISPs who are looking to understand how well NetFlix (and other services) are working on their networks in an attempt to improve service to their customers. These are small-ish ISPs, and I'm trying to help them move from guessing how well their customers are working based on high level traffic graphs to something more reliable and robust. I'd love to chat about this if you're available, email is in my profile.
It is a feature. When my kids (or parents) are watching and the stream suffers, I'd rather it step down the bitrate momentarily than cut the stream or pause to buffer.
Remember realnetworks and the constant "buffering.. buffering..". I'm glad to leave that in the past.
This highlights the conflict of interest that cable ISPs have. If they provide you good internet service, by either
allowing embedded Netflix (and Google, Amazon, etc) cache appliances into their network, or had add more downstream bandwidth, then you might have cut the cord. However, by giving you crappy internet service, they have retained you as a cable customer.
My service is fine for all other use cases but remote HD video, with its need for constant high bandwidth, is always going to be a fundamentally more difficult problem to solve than local digital video, traditional cable-delivered video, or regular downloads that can tolerate variance in transfer rate. It doesn't feel like it takes a conspiracy to explain that. When cable companies wanted to be our email or news/content portals I never had any issues using competing services on cable internet.
I was never going to cut the cord. I like having many channels and the serendipity of a wide variety of programming that's on "now" to choose from. There are shows/movies I never would have chosen myself but I stumbled onto them because they were "on" and they became favorites.
With a DVR I can also time-shift so I'm not tied to watching something on the broadcasters' schedule when I don't want to.
I tried Netflix to augment this with what I thought would be a big influx of other programming but it wasn't only the streaming experience I didn't like, I was unimpressed with the selection of the Netflix library. If they do another season of Arrested Development, I'll subscribe for a few months to support that.
I like the idea of getting ISPs into internal conflict: the folks responsible for making sure that speed checks like speedtest.net run quickly will be fighting the folks responsible for throttling Netflix.
But, I think the throttling folks will ultimately win. In that case, I guess Netflix is laying out a good case for consumers to complain, so it's win-win.
Really nice and easy to use -- the test starts way quicker than speedtest.net.
However, am I missing something, or does this only test downloading? I guess that makes sense for Netflix's use case, but I'm usually at least as interested in knowing my upload speed, because with typical asymmetric connections that can be a bigger bottleneck for video calls and content-production workloads.
>Fast.com estimates your current download speed. You will generally be able to get this speed from leading internet services, which use globally distributed servers.
>Why does Fast.com only report on download speed?
>Download speed is most relevant for people who are consuming content on the Internet, and we want fast.com to be very simple and fast.
Well, the nice thing about this is that they test with real-world streaming data that ISPs can't game like the do on SpeedTest.net. What's the upload equivalent of that? Flickr uploads?
All very offtopic, but uploads on the Flickr site usually timeout on my home broadband these days. I fear they've been tweaking their servers for classier users than I am.
Very interesting, and it confirmed my suspicions that my ISP throttles me (or at least, tries to).
I'm using Numericable from Paris and got 18Mbps to Netflix, 40Mbps to their comparison test. By going through an SSH tunnel (which makes a 230km detour through Roubaix), I get 39Mbps to both Netflix and control.
I am rather surprised that the bandwith loss caused by the SSH tunnel is so small.
It's also possible that the route between Numericable and Netflix is insufficiently provisioned, unlike the route from Numericable to your SSH server to Netflix, which can have more capacity.
What I mean is that's probably not an intentional throttling. Anyway, it's still a bad service. I have a similar issue sometimes with SFR...
Is there a place to share these results? I live in northern New Jersey with Verizon and get 55-58 for every speed test I can find. Netflix seems to be faster if anything, but I'm guessing that's due to the competition with Time Warner.
Oh man this is awesome. I can't wait till people start calling thier ISPs claiming they aren't getting the speeds they pay for, only for the poor agent to have to explain how peering agreements work.
I'm about to leave an ISP where many/most of the techs know what a peering agreement is and are happy to help troubleshoot tricky issues (e.g. did you know PJSIP doesn't do NAT rewriting?)
It's sad really, but I can now get similar quality of service for half the cost so I'd be silly to keep paying what I am.
It makes sense. Comcast is directly paid off (I mean peered) with Netflix, and they overprovision intentionally. See https://www.dslreports.com/faq/15643 if you are curious.
That link (https://www.dslreports.com/faq/15643) seems to be a list of Comcast's available bandwidth speeds, and doesn't mention Netflix or peering. Did you intend to post something that describes the interconnect agreement between the two companies?
Couldn't that just be gamed by allowing fast speeds for the initial connection of X size or Y Time and then dropping down to less once they guesstimate the test is over?
That proposed plan could probably be reverse gamed by Netflix by buffering streamed content with multiple separate connections. Or by including a way to check the speed mid stream. Or by allowing extended tests on fast.com. This is a classic cat and mouse game.
I'd bet this is a move to make the ISPs that are throttling them look bad. If people start to use it to check their speeds and they are downloading Netflix content from Netflix and the ISP is throttling, it will look slower than it is and more people will likely complain.
Why is it evil? And if the ISP is throttling Netflix, it doesn't "look slower than it is", it just accurately reports the throttled speed.
I mean sure, it's definitely to make ISPs look bad... if they're throttling Netflix connections or otherwise providing slower than advertised speeds. Which doesn't sound terribly evil to me.
That was what I meant. It isn't really evil. "Suitably evil", in my mind, implies that they are responding with a suitable amount of evil to what they've been faced with.
OP looks like he's calling Netflix's actions evil, and the post you're replying to is questioning that. I think everybody in this comment chain agrees that throttling based on domain is illegal and immoral for the ISP to do.
For the record, I meant evil in a fun way. Like a cartoon villain. Suitably meaning that it was a good equal response to what they've been dealing with.
It very much depends on where the local speedtest.net server is in relation to you in terms of network topology. If you're in a large city like SF or Seattle, your ISP might host a 1U box from ookla or directly peer with them at a major IX point less than 15 milliseconds away from you.
For me, it's getting stuff from https://*.cogeco.isp.nflxvideo.net -- which indicates my ISP (Cogeco) is part of their Open Connect [1] program with an on-network netflix cache.
Other people are reporting downloads from https://*.ix.nflxvideo.net, which appears to be the Netflix cloud infrastructure.
It downloads data from 5 URLs every time, but their sizes fluctuate, something like ~25MB, ~25MB, ~20MB, ~2.2MB, ~1.2MB.
The contents of each response appears to be the same (though truncated at a difference place), with the beginning starting with:
5d b9 3c a9 c3 b4 20 30 b9 bc 47 06 ab 63 22 11
`file` doesn't recognize what this is.
----
Since it's https, ISPs shouldn't be able to easily game this (eg: make this go fast, but still throttle video content).
So one potential way would be to only start throttling after 25MB is downloaded (or after a connection is open for ~2 minutes): does anyone know how Netflix actually streams? If they have separate HTTP sessions for 'chunks' of a video, then presumably this wouldn't work.
They could see if a user visits fast.com and then unthrottle for some amount of time. I'm not sure if ISPs have the infrastructure to do a complex rule like this though (anyone know?). I also think this would be relatively easy for users to notice (anytime they visit fast.com, their netflix problems disappear for a while) and there would be a pretty big backlash about something so blatant.
Odd, on a dedicated 500 mbit line I've now gotten 6 different results, ranging from 350-500. Speedtest.net indicates a stable 500+ mbit line, downloads from very fast servers always max it out at 500 as well.
Besides stabilizing it a bit, getting the upload on there would be amazing, it's certainly a lot nicer for the eye than speedtest.net.
That's because ISPs game SpeedTest.net and give you your full bandwidth for the test data. So the 350-500 is more accurate to your actual milage (at least for Netflix's peering agreements).
That would be totally valid, except I get a perfectly fine and stable 500 down from, for example, Microsoft's MSDN, depending on the size of the ISO showing this straight for a couple of minutes..
Right, this tool only tests your ISP's speed for Netflix content, which most tend to throttle. For the majority of people, Netflix is the only way that the saturate their bandwidth, so this tool is more relevant to them than to people downloading ISOs. I generally speed test by wget'ing a large ISO because those numbers are more relevant to me.
Different paths will have different amounts of bandwidth and latency, especially at different times when some parts of the network are heavily loaded.
In certain cases (ie: transatlantic cables), there is only one quick path between US and Europe. If that cable is heavily loaded, then your bandwidth when talking to a European server will be slower.
We are seeing about a 10% swing on multiple tests. Close; typical network usage. It would be interesting to run this on Friday night about 8:30PM EST. We normally see a network slowdown then BECAUSE of netflix....
As said in other comments, this is because speedtest.net largely tests within your isps' network, not the path off their network to other ISPs or to netflix/netflix like services.
To put this another way, speedtest is kind of lying to you in favor of your isp, which is the source of their funding.
This tool exists exactly to highlight cases like yours.
Sounds like your ISP has over-subscribed its connection to whatever IX is serving this to you, and that they're hosting speedtest locally in their network.
I'm a Comcast customer. When I run speedtest & tcpdump the traffic, i'm nearly always talking to something .comcast.net, and it is just a few hops away via traceroute.
When I use fast.com, I'm talking to various ix.nflxvideo.net hosts, all at least 4 to 6 hops farther away.
The amusing thing is that I get better speed with the fast,com (140Mb/s) than with speedtest (133Mb/s).
Interesting. I'm on AT&T Gigabit (Google Fiber competitor here in Austin) and I got 660Mbit on fast.com (Windows 10 desktop directly connected via Ethernet.) A MacBook Pro laptop with Wifi (802.11ac) connected a few rooms away shows 260Mbit.
Are you wireless or wired? If wired, that seems quite low.
Thanks for reporting. I'll test when I get home, but I'm also in Austin and have Gigapower. I was afraid that it would almost embarrassingly low, but your results gave me hope. I'm curious what the Google Fiber and Grande users here get.
I'm on Grande's gigabit and am getting ~180-200 Mbps over 802.11ac a couple rooms away. Sadly I don't have an ethernet dongle for my macbook pro to test a wired connection.
I'm also on AT&T Gigabit in Austin, wired, reporting only 490Mbps on fast.com. However, I can download from Steam at 950Mbps, and report similar speeds on other tests.
I get 450Mbps on centurylink gigabit fiber in Seattle. I get 900+ Mbps via speedtest.net. I am hardwired direct to modem on gigabit LAN. I am happy to have this problem, but I have noticed most speedtests (depending on endpoint) don't report gigabit connection speeds well. I am not sure if there is an upper limit to this Netflix test or if I am actually getting throttled.
Not to sound ignorant, but what's the point? Why would Netflix go through the trouble of acquiring what I suspect to be a fairly expensive domain just to show how fast one's internet speed is?
> Why would Netflix go through the trouble of acquiring what I suspect to be a fairly expensive domain just to show how fast one's internet speed is?
Fast.com shows how fast your connection to Netflix's servers are.
Given how popular a use of the internet Netflix is (and that using it for Netflix is one of the reasons people pay for particular internet speeds), and given that many ISPs provide connections which, for various reasons, provide less speed on connections to Netflix than users might expect from advertised speeds, this seems to be useful both to consumers and Netflix. The motives clearly are not altruistic, even though there is consumer utility.
Probaby going to annoy the heck out of a lot of ISPs...
> I find it hard to believe their motives are purely altruistic.
Absolutely selfish. Their goal is to make boatloads of money by pressuring ISPs to allow their customers to access Netflix at the speeds the customers paid for. Once customers can use their high speed internet, many will choose to be Netflix customers.
It's interesting to describe something that gets people what they pay for as selfish on the part of a third party. I mean, yes, there is something in it for netflix, but it arises from making things better for their users. I'd describe that as win-win, personally.
Now, killing performance on a service your users want so you can force an inferior product on them? That's selfish.
> It's interesting to describe something that gets people what they pay for as selfish on the part of a third party.
It's selfish because it is done with the goal of increasing Netflix's revenue.
Now, in this case selfish doesn't mean evil. To the contrary, Netflix's success will mean happier people, more satisfied customers. Much in the same way auto manufacturers make safer cars for higher profits, or surgeons can make a comfortable living saving lives.
Netflix is simply trying to make a buck, and if some people get helped in the process, that's just collateral benefit.
>Now, in this case selfish doesn't mean evil. To the contrary, Netflix's success will mean happier people, more satisfied customers. Much in the same way auto manufacturers make safer cars for higher profits, or surgeons can make a comfortable living saving lives.
I would not call that "selfish", nor would I call it "altruistic". I would perhaps say "fair" or "just".
"Selfish" is generally a negative word. We often refer to any self-regarding behavior as selfish: both rational self-interest pursued in a manner respecting the needs of others, and the miserliness or enviousness that (often deliberately) pursues self-interest at the expense of others needs. Sometimes we state the equivocation in the form of a paradox to catch our audience's attention (the capitalist's aphorism "greed is good") and sometimes as a slur ("those greedy capitalists!"). However, I think it would be helpful if in the normal case we recognized that not all self-regarding behavior is selfish.
Note: I don't mean to nitpick, and I agree with the spirit of your post. Rather, your words were to me a catalyst for some thoughts on moral discourse.
If this (the concept/reasons behind fast.com) gets promoted correctly to (non-techie) consumers, then this could serve to get a conversation going at least as it pertains to consumer-centric concerns such as net neutrality, non-fraud business tactics (Dear ISP, I pay you X money so I expect Y level of service), etc. Sure, net neutrality and fraud are not new topics...but within the technology space, they're often difficult for consumers to wrap their brain around, and of course not in the best position to call out their providers (whenever bad behavior occurs). Simple demonstrations like this help to arm and inform the regular Joe/Jane to make better decisions related to how to spend their hard-earned money.
Although I'll admit that I'm a biased fan of netflix, kudos to them for doing this!
Considering Netflix accounts for 37% of all downstream traffic on the entire internet during peak hours it seems that the investment would be well worth the cost of the domain if they are having trouble with ISP's limiting users speeds to their servers (and we know they are!). Now instead of ISP's gaming speedtest.net by giving it priority speeds, Netflix gives users something to test and view the bandwidth being served to their streaming servers.
This is nice and great that it loads quickly with no bloat or distractions. Not sure about the domain name though, as it's not immediately obvious what the site is for.
While cool, I can't believe they bought and use fast.com for something so simple. Fast.com has to be worth some coin. Anybody have any idea what that domain is worth?
I'm pretty sure Google is about to release a speed test tool embedded directly into its SRP for speed-test-related queries.
Similarly to how they eliminated the need for third-party IP address checking tools by returning your actual IP address when you search for "what's my ip address".
They didn't eliminate the need. To get my IP from there programmatically, I'd have to parse the DOM. Or, I can "curl https://i.ngx.cc", and have it immediately available.
The amount of data netflix will collect from this is exciting! I can only imagine the stories it will tell once hundreds of thousands of people use it. It would be fantastic to see how the agreements between ISPs and netflix affect the data transfer rates.
The secret here is that this is measuring speed to Netflix, which on many ISPs is throttled well beyond what other speed tests will report. I'm in the Boston area and getting 100Mbps on fast.com, but getting 250Mbps on speedtest.net. Fast.com is even encouraging you to compare their result with speedtest.net.
It's a pretty smart tactic. If you're paying for 150Mbps, but being throttled to 100Mbps on Netflix and your ISP doesn't tell you, you're not really getting what you pay for.
It definitely is.. I have a newish Samsung smart TV and the streaming quality was always garbage. I got an Apple TV and HBO and Netflix consistently stream at their highest bitrates via the same 150mbps internet connection.
My office also uses Comcast/xfinity for business. Incredibly, in 2016, we had to get a $40 device that reboots the modem and router every single day once a day when the office is closed. This is the solution we were given to maintain decent speeds with their hardware. And it works.
OK, so I get 48 on fast.com and decided to use the link to compare on speedtest.net. There I get 101 down, 112 up.
So while 48 seems very fast to me (I get 19 at work) it's a lot less than 101. Is Verizon throttling the connection or is NetFlix not giving me more the ~50? At what point is the cap on NetFlix's side and not the client connection?
OK, I get that. My point is at what point is a connection fast enough that NF won't serve you any faster? I get 101 download on speedtest but I've never downloaded an ISO file and had it download any faster than 5-6. The service on the other end has a say in how fast my downloads are. So my question is there some theoretical cap on the NF side that says anything over (arbitrary-number> is fast enough so clients are capped at that.
I'm on FIOS so I would not be surprised at all that VZ throttles the connection but on the other hand why cap it at ~50 and not at 5 or whatever? At 50 my connection is fast enough for me to watch full HD so it's not as if there turning me into a frustrated customer who decides cutting the cable is more hassle than it's worth...
One thing that Fast.com doesn't say is the top speed that Netflix's servers can deliver. I have 250 Mb fiber connection, and Fast.com tops out at about 150 Mbps for me. Speedtest.net consistently shows 250-280 Mbps. But it's unclear whether the lower rate is Netflix's servers limit, or my ISP throttling them.
I am on a MacBook Air Early 2014 and my current link speed is 144 Mbit/s according to Network Utility, but fast.com shows between 210 to 230 Mbps on each run.
Speedtest.net results are consistent as before at ~38 Mpbs, which is what I would expect from the routers around me.
The amount of data you upload to NetFlix to watch a stream is negligible so really doesn't apply in this case. This site is strictly to see if your ISP is purposely slowing down your streaming specifically from NetFlix.
My internet speed (according to fast.com) is 0. Adblock & uBlock off on the site & fast.com uses https. Not sure why it wouldn't be working, no VPN in middle. Anyone else having issues?
edit: Speedtest.net was ~38mbps down. Is a netflix subscription nec. for this?
I can reach netflix.com. I can now reach fast.com. I am skeptical that the results are correct, i pay for 50mbps, typcally receive ~40ish, and fast.com says avg of 56
edit: can't reach servers again. no idea why, currently behind vpn now. No disruptions anywhere. I wonder if the site is just under heavy load and failing but not returning proper status codes.
The download speed is absolutely incorrect, I live in a 3rd world country and have a 2Mb connection I get max 200/kb max download rate, fast.com is showing a download speed of 1.2Mb I really wish it was true
>Why doesn't netflix just try to by-pass ISPs by rolling out their own service?
Because it's hugely expensive. Fios cost Verizon ~30 billion dollars and it only covers a fraction of the USA and large parts of it are already outdated.
Speedtest is a more accurate measure of what you're actually paying for - which is a connection to the ATT network. The speedtest server is right on their network most likely.
Fast.com is a more functional test, specifically measuring how fast a connection on the ATT network in that area to the Netflix servers is. You're not paying ATT for a 24mb/s connection to every server on the entire internet because that would be impossible (how could your ISP promise you that much bandwidth across a transatlantic cable, or over a satellite link, or to a server connected via dialup). More generally, you're testing the strength of their peering and the route available for a specific purpose (watching Netflix).
They aren't measuring the same thing at all. It would be like buying a 100mb/s router, plugging it into 20mb/s internet and then complaining that your router is broken because you can't download at 100mb/s. Your router will still show your ability to contact other computers in your house as 100mb/s (analogous to speedtest), but reaching computers outside your network (analogous to fast) is limited by the speed of your connection to the next larger network.
Speedtest.net provides many different servers, most of which are not on the AT&T network. The IP is not in the AT&T space and the AS path leaves AT&Ts network.
In the end until someone can provide a dump of what is in the routers settings it is all a guessing game.
I frustratingly pay the same amount while other ISPs in southern California offer 300 Mb/s for the same price. My apartment isn't wired for anything else though which is a bummer.
How could they? It is traffic going back and forth to the Netflix servers over https. The ISP can't tell WHAT is in that traffic, only where it is going. Thus, if they throttle Netflix, they throttle fast.com.
The simplest way would be to cheat and prioritize traffic for a few minutes and then throttle. That way, any speed test would come out fine but not actual shows.
As others have said, the simple way would be to optimize for a time period. However, another way would be to monitor the site and watch the data patterns that emerge. using those data patterns, the ISPs will be able to have a better idea of how to game the test.
Not OP but it's a large, hideous site with a lot of junk I don't care about. The faux-futuristic map with an arc to a pyramid? The colored odometer with a variable scale? Creating an account? Ads?
As others have mentioned, ISPs can game the speedtest.net results, whereas with fast.com users will get exactly what they want: a single number that shows how fast they get Netflix. No frills, nothing else. Some of the advanced features of speedtest.net are useful for some advanced users, but it's a weird de-facto standard that caters to the tiny minority.
Those are pretty good points. The big appeal for fast.com for me is that it measures the connection to Netflix's services and ISP's can't game it, but I still wish it gave more information than a single number.
All I want from speedtest.net is two numbers: up and down. That is everything fast.com gives me, and nothing more. I don't care about the fancy/weird laptop battleship image, ads, etc.
Also the "powered by ookla" is the weirdest branding I've ever seen. Not only does it bother me, but makes me trust the site way less. I didn't know what ookla was until I googled it just now.
Redesign/rebrand speedtest.net so it matches ookla. Right now speedtest looks like a scam, ookla looks official.
My ISP runs speedtest.net in my local area. For that reason alone I don't believe it. My speeds displayed on speedtest.net are significantly higher than other speed test websites.
Wow that design on that site is a lot better. I've also never actually seen a big website host ads on their own servers, and still not get blocked by abp+. Ads also don't have a redirect, and they are adsense. Interesting.
Also this makes me feel weird: (all the links still go to c.speedtest but have google's content)
is there a bandwidth cap? i constantly get 160Mbps but i know for sure that our office line can do way more. speedtest.net is always close to 900Mbps. maybe speedtest.net has a endpoint within the ISP backbone and netflix not? or is the peering between AWS and my ISP?
Yes it does. For example here for the speedtest it started a connection to: ipv4_1-lagg0-c002.1.hmo001.telmex.isp.nflxvideo.net (which is my ISP Telmex in Hermosillo) and ipv4_1-cxl0-c041.1.dfw003.ix.nflxvideo.net which is in Dallas.
I'm on a basic 1 GBps home connection here in Japan. Costs $35/mo. My ISP is kinda crappy so I usually only get around 300-500 MBit - I've been looking into changing ISPs (the fibre network has open competition)
You're right, uplink is pretty much useless if you're only consuming Netflix. I guess I'm still infected by 90s-era ideals about the internet being a place for the free exchange of information, where any tech-savvy person should be able to run their own server, etc, etc. I'd at least like to be able to upload photos of my son without my connection slowing to a crawl.
I have had that conversation with developers, where I literally was pointing them at the Wikipedia article and quoting bits back at them. Fun times.[citation needed]
dude. the guy owns slow.com, it's his traffic. i'm talking about OWNERSHIP, not the http protocol. please read the original post about 'stealing' traffic.
he owns the domain name, he gets to do what he wants with it. _that's how dns works_.
i don't know why the hell redirects even got brought up.
this entire thread is pedants making shit up so they can be pedantic while ignoring the actual conversation.
uh, okay. the "dn" in "dns" means "domain name", which according to my understanding of the basics of the internet works, is what fast.com and slow.com are.
This is important because unlike your average Internet speed test (which ISPs take pains to optimize), there's a very real possibility that your ISP is happy to let your Netflix experience suffer - assuming they don't throttle it outright - as previously mentioned on HN:
https://news.ycombinator.com/item?id=7183682
https://news.ycombinator.com/item?id=7853603