Ask Slashdot: An Accurate Broadband Speed Test?

First time accepted submitter kyrcant writes Is there a way to accurately really test my "broadband" connection? I don't trust the usual sites, the first ones I found via Google. I suspect (and found) that at least some of them are directly affiliated with ISPs, and I further suspect that traffic to those addresses is probably prioritized, so people will think they're getting a good deal. The speeds I experience are much, much slower than the speed tests show I'm capable of. For a while I thought it might be the sites themselves, but they load faster on my T-Mobile HTC One via 4G than on my laptop via WiFi through a cable modem connection. Is there a speed test site that has a variable or untraceable IP address, so that the traffic can't be prioritized by my ISP (call them "ConCazt")? If not, which sites are not in any way affiliated with ISPs? Is there a way to test it using YouTube or downloading a set file which can be compared to other users' results? (Have your own question for the teeming masses? Ask away — be sure to include appropriate detail and context — via the Slashdot submission form.)

Comments

None

[AK Marc]

They are all ISP run, or open to bribery. The most independent one I've seen is https://www.google.com/get/vid... which is an ISP quality measure, not a speedtest.

Re:None

[malakai]

I like http://speedof.me/

It's fast, works with HTML5, works on mobile, tablet, desktop. As far as I can tell, it's hosted in the Amazon Cloud.

-frank

Re:None

[peted56]

Not bad, http://speedof.me/show.php?img...

Re:None

[gweilo8888]

Why would an ISP bribe them, when all they need do is identify when a speed test is occurring and give that traffic priority over all else? They wouldn't, and your tinfoil hat is too tight if you think otherwise.

Speed test sites are fundamentally flawed, but bribery has nothing to do with it.

Re:None

[AK Marc]

I work for an ISP, and I've been solicited for bribes. It's not tinfoil, it's reality.

Re:None

[fafalone]

That site reported my downstream as 40mbps (close)... but pegged my upload at 7mbps. My downstream is advertised as 50, but I see 58 in my tests. But my upstream is also 50, and I get 50 in my tests. Why it's so grossly off I don't know; it says the server is very close to me. But a lot of speed tests are off. My testing method is the real world: either start several torrents, or open multiple downloads from file locker sites (which, btw, advertise 'uncapped speeds', but cap per-file... so I can saturate my connection but it takes 6-7 concurrent transfers from the same server). Single-connection transfers have saturated my line at 58/50, but it's fairly rare.

Re:None

[mjwx]

They are all ISP run, or open to bribery.

Not quite.

What you do is download a file of a set size from a server you own and control on your ISP's network. I do this with site-to-site links. I have a 20, 200 and 500 MB file of randomness that I copy, the time taken to download is a pretty good indication of the links aggregate speed.

But no, I wouldn't trust sites like Speedtest to be accurate because ISPs that are dishonest will just prioritise traffic going to that site.

With residential broadband, the only thing you can do is sign up with an ISP that has a good reputation and doesn't lock you into a X month contract (so you can leave at any time). Australia has a few good ones like iinet and Amcom, but in the good ol' US you're pretty much stuffed because of local monopolies.

Re:None

[AK Marc]

What you do is download a file of a set size from a server you own and control on your ISP's network. I do this with site-to-site links. I have a 20, 200 and 500 MB file of randomness that I copy, the time taken to download is a pretty good indication of the links aggregate speed.

That worked in '90s DSL world, where they'd oversubscribe the links from the CO to the core. But now, the links are not oversubscribed from the CO to the core, but are oversubscribed from the core to the Internet. So you'll get better results on your test than you'd be able to get a file from Microsoft. I remember the days in the early '90s where I'd get files from MS at local interface speeds, so long as I had a T1 or other direct connection, rather than "shared" cable or DSL.

Re:None

[mjwx]

That worked in '90s DSL world, where they'd oversubscribe the links from the CO to the core. But now, the links are not oversubscribed from the CO to the core,

Not in my country, the last mile is still the slowest part.

But when you're paying for an SLA for a site to site link, this does not matter (in fact, my traffic should not even leave the ISP's network).

Re:None

[Bert64]

Depending on latency, buffer sizes and tcp window scaling etc, a single connection is usually much slower than multiple connections used by torrents...
Depending on the window scaling settings at either end, your machine may only process up to 64kb of data (per tcp connection) before it needs to send an acknowledgement, which can result in very poor performance over high latency and/or long distance links.

Re:None

[Bengie]

But now, the links are not oversubscribed from the CO to the core, but are oversubscribed from the core to the Internet.

Not my ISP. I was talking to a Sr Network Admin who had been working at my ISP over 15 years and he said they try to keep their trunk below 50% utilization during peak hours. They way he described it is that the link is a teamed logical connection, and if one of the links goes bad and 50% of the bandwidth drops out, there should still be enough bandwidth that the customers won't notice.

Their total bandwidth, including failovers and backups, is about 6x peak usage.

The biggest issue my ISP has is there are technically "small", and sometimes gets a DDOS against a business customer that saturates their trunk, but it's not often.

Re:None

[pixelpusher220]

Not saying it doesn't happen, the point is it isn't necessary. If I'm an ISP, prioritizing traffic from known speedtest sites is a lot easier, cheaper and covers more people.

But to the point, OP asked:

Why would an ISP bribe [a speed test site]

And you answered

I work for an ISP, and I've been solicited for bribes

That's the exact reverse situation he described...

Re:None

[AK Marc]

Why would an ISP pay a bribe? Because the speed test is used by competitors to "prove" the ISP has bad service, and the only way to improve it (possibly because the people running the "independent" speed test are throttling it for those who don't pay the extortion), is to pay a bribe.

Gotta love the "free market".

Re:None

[AK Marc]

but you said worked at an ISP and have been solicited for bribes; i.e. been offered money by the speed test companies.

i.e. I was asked to pay a bribe. I was solicited. I was approached by someone and asked to pay a bribe.

If I were offered money by speed test companies, I would be offered a bribe, not being solicited for one.

If you're claiming the speed test companies tried to get you to pay them for better results, that's extortion, not a bribe.

Being solicited for a bribe is being extorted. http://www.bullmanlawgroup.com... The cop extorted someone, and it's called "solicited a bribe". Yes, I chose wording you wouldn't, but that doesn't make it wrong.

Really?

[Maxwell]

How are people not aware of DSLReports and their speed tests? And how could this possibly make /.?

Also, your wi-fi sucks. Get a cable if you want to know what your real speed is.

Re:Really?

[Anonymous Coward]

Really. This shouldn't be here but there are also torrents (they seem to be able to saturate my 15Mb connection easily if somewhat popular). You could also download a windows iso from microsofts CDN - they seem to have decent bandwidth.

Re:Really?

[Anonymous Coward]

"Get a cable if you want to know what your real speed is"

that would be the least problematic way of being sure it isn't your wifi acting up though, wouldn't it?

Re:Really?

[AaronLS]

Agreed he should accuse WIFI with no evidence, but at the same time it is not a legitimize test with WIFI in the loop. If he's experience connection issues or measuring performance of his cable connection, then he should do a direct connection to eliminate WIFI since it is very susceptible to many issues that could affect performance. Only then can he point fingers at the cable connection.

Re:Really?

[AaronLS]

Oh I should have proof read that, there is some grammar murdering going on there. You get the idea though.

Re:Really?

[dotwhynot]

Not really fair to immediately disregard the quality of the WiFi connection. It could be well in excess of the ISP connection.

I have a 40/20 mbps broadband, and independent local non-ISP speed tests give the same result on WiFi as ethernet, around 37-38/18-19 mbps. But, I do agree that if you get shitty results, you should try to rule out that shitty WiFi is the reason.

Re:Really?

[synapse7]

So which speed test from DSLReports do you suggest we use?

Re:Really?

[JeffAtl]

What should he accuse WIFI of doing?

Re:Really?

[Maxwell]

Maybe you should follow the instructions on their Tools page:

Speed Test
We have the following speed tests

Flash (Adobe) download/upload speed test
Accurate for tests of residential DSL and cable connections

Java download/upload speed test
Capable of higher speed testing, for example, fiber

Mobile browser Speed and Latency Test (http://i.dslr.net/iphone_speedtest.html)
Javascript Speedtest, for mobile full featured browsers (iPhone, Android and so on)

Re:Really?

[afidel]

Microsoft uses Akamai, and unless you use their download manager the throughput is kinda crap, at least on AT&T with our own DNS servers.

Re:Really?

[sexconker]

Not really fair to immediately disregard the quality of the WiFi connection. It could be well in excess of the ISP connection.

Have you ever even used wifi?
802.11b = 11 mbps, real world throughput = ~6 mbps.
802.11g = 54 mbps, real world throughput = ~20 mbps.
802.11n = 300 mbps, real world throughput = ~60 mbps.
802.11ac = 1500 mbps, real world throughput = ~160 mbps.

Re:Really?

[StingyJack]

Killing Laura Palmer

Re:Really?

[lgw]

My wifi with the EM clutter from 20 visible Wifi networks, as measured from across my apartment: dial-up speed on some days, around 100 mbps on others. I wouldn't use it for an ISP speed test.

Re:Really?

[fahrbot-bot]

Oh I should have proof read that, there is some grammar murdering going on there. You get the idea though.

Don't worry, we just assumed your grammar got messed up over your WiFi connection.

Re:Really?

[Eunuchswear]

Not really fair to immediately disregard the quality of the WiFi connection. It could be well in excess of the ISP connection.

Then the answer to your speedtest is: "your ISP is crap".

Re:Really?

[gweilo8888]

What about DSLReports' speed tests makes them inherently more trustworthy than any other? All the ISP needs to do is prioritize speed test traffic over all other traffic, and speed tests become meaningless. Heck, they could even raise your cap for the duration of a speed test to make it seem like you got a faster speed while actually capping you to a lower one. The whole thing is a joke.

Re:Really?

[jimbolauski]

20 networks, you should probably try 5GHz WiFi, it's supposed to utilize dynamic frequency selection and dynamic power control and fewer people use it. Even if 5Ghz was widely adopted around you it "should" only slow down if everyone is using it at the same time and some of those people are further away then you are to their wireless router. You will have less range then 2.5GHz but in a crowded apartment complex that is a good thing because there is less chance of your neighbor crapping on your SNR.

Re:Really?

[lgw]

it "should" only slow down if everyone is using it at the same time and some of those people are further away then you are to their wireless router.

That's pretty much the case with 15' wide, 40' long townhouses all in a row, where everyone has modern WiFi routers (and there's also a row of storefronts along the back of the townhouses, and some businesses there offering free WiFi to customers, to further crowd the area). I just use long ethernet cables for the important endpoints.

Re:Really?

[TemporalBeing]

How are people not aware of DSLReports and their speed tests? And how could this possibly make /.?

Also, your wi-fi sucks. Get a cable if you want to know what your real speed is.

The ISPs cheat for the speed tests by temporarily increasing your bandwidth so that the tests detect a higher transfer rate than what they are actually giving you. They don't even prioritize just the DSL testing sites either; at least AT&T DSL doesn't.

Re:Really?

[mattack2]

temporarily increasing your bandwidth

So that's different from what they actually advertise, speeding up the first X megabytes or whatever? (I _think_ that's what Xfinity Blast is, but I can't actually find a description on their site at the moment.)

Re: Really?

[theqmann]

I've recorded sustained 700-800 mbps to a local networked fileserver over an 802.11ac connection. The fileserver's probably the limiting factor in this too.

Re:Really?

[madbrain]

Your numbers reflect my experience as well except for 802.11ac .

I am getting between 300 to 400 mbps real world throughput depending on how far the wifi device is from the router.
That's with connection rates between 500 and 1300 mbps - the later achieved with an Asus PCIE 802.11ac NIC.

IMO, 802.11ac is a huge advance over 802.11n.

Re:Really?

[AK Marc]

Maybe it was a comment about all the tests being in North America, and many people not living near one of those servers. The trans-Atlantic performance of the speed test isn't amazing.

Re:Really?

[dbIII]

Sadly the new Australian definition of "broadband" appears to be lower than any of those either way, let along a nerfed ADSL upload.

Re: Really?

[sexconker]

That's factually incorrect.

Real world tests of 802.11n/ac show that they deliver much closer to the headline rate than any of the older protocols.
802.11n can actually go up to 450Mbps with an actual local transfer rate of around 350, while 802.11ac wave 1 can get even closer to the headline rate. Depending on the transfer direction good WiFi can flood a 100mbit duplex link.

The key issue is with spectrum efficiency is utilisation. Those numbers are 1 device and 1 radio. As soon as you add another device, you're sharing that spectrum (until wave 2 with MU-MIMO complicates this). You also get more signalling overhead and collision retries with each ssid, ap, client in the same audible space.

I've run both n and ac over a dozen different networks and I the only time I can ever get anywhere NEAR rated speeds is if the AP and the client are in a basement. Wifi degrades the instant you bring it out of a testing cave, and it turns to absolute shit the instant your neighbor fires up his own radio of the same class. I don't give a flying FUCK about theoretical rates in a best-case scenario, I care about how it's going to work in a typical deployment. Wifi is, and continues to be, horribly designed with regards to channel spacing. And the "solutions" to that have been higher frequency and lower power, trading range for speed.

Re:Really?

[sexconker]

Your numbers reflect my experience as well except for 802.11ac .

I am getting between 300 to 400 mbps real world throughput depending on how far the wifi device is from the router.
That's with connection rates between 500 and 1300 mbps - the later achieved with an Asus PCIE 802.11ac NIC.

IMO, 802.11ac is a huge advance over 802.11n.

Just wait until your neighbors pick up an ac router.

Re:Really?

[RockDoctor]

That's pretty much the case with 15' wide, 40' long townhouses all in a row, where everyone has modern WiFi routers

Normal suburban street lined with 2/3/4 bedroom houses with one garage space per house ; I'm seeing 14 WiFi networks plus my own one. Without getting out a survey tape, I'd say that's essentially every house within 50m radius, which sounds about right (brick and tile construction, little steelwork apart from garage doors). In denser, multi-story city centre accommodation, 40 networks within reach is easily achievable.

Speedtest

[nitsew]

You can download the speedtest widget, and load it on a webserver, and then use that to test your speed.

http://www.speedtest.net/mini....

Re:Speedtest

[InvisiBill]

You can download the speedtest widget, and load it on a webserver, and then use that to test your speed.

http://www.speedtest.net/mini....

If you have a server you can install it on, Speedtest Mini is great. It uses their same basic setup, but allows you to run it somewhere other than a standard "speedtest" server, in case you think those servers are being handled differently.

VPS

[ledow]

Rent or trial a VPS. You can get them for literally a few pounds/dollars per month.

Put a large file on Apache on it.

Download the file from several places.

Rename the file on the server to check it's not cached.

The "upper limit" on this is then the VPS, which generally are connected direct to 100mbps lines in a datacenter somewhere. If you think it's limited by the VPS, get another from another provider. Or load up iptraf or some packet capture and see how it did.

Speedtest websites are indicative only, and are cheated on by some places. Your own website can't be cheated on - you will see the request coming in and can watch the outgoing traffic to see where the bottleneck lies.

Re:VPS

[TubeSteak]

Rent or trial a VPS. You can get them for literally a few pounds/dollars per month.

The old timey way of doing speed tests is to hit up FTPs and see what your max sustained speed is.

Re:VPS

[Anonymous Coward]

That's almost good advice.
But keep in mind that this could still be susceptible to traffic prioritization of TCP port 80. Which might not be a problem, unless you're actually wanting to use the Internet for something else (HTTPS, torrents, or other).
One thing you could do is to have your test web server listen in a non-standard port. That might fool some prioritization techniques, and show you the raw speed.
Another thing you could do is to actually test whatever type of traffic is important to you. That might not be web traffic at all. If you care about low latency for a game, then web traffic might be a very low priority. What is most important to you will entirely depend on what your needs/desires are.
My final tidbit to add to this: trust not in advertised speeds. If you don't have a contract with a specific service level agreement, then you're almost certain to have no guarantee that your average speed will actually match what is advertised. If, on the other hand, you do have a service level agreement, then you can only be assured that they will provide enough speed to reduce their likelihood of losing a lawsuit, and even that might only be true if they expect that the lawsuit would cost them more than it costs them to deliver what they agreed to.
I was going to make a snide remark that if you really want what you are promised, go with dial-up. But then I remembered that even my old 28.8 dial-up modem typically connected at a speed slower than the maximum speed advertised.
See, the most certain way to have a guaranteed maximum speed is to place an artificial limit, because usually technology is granting us some maximum speeds that are really only feasible some of the time. So if we can get 300xbps regularly and 400xbps on occasion, the safe way to guarantee is to limit the connections to 300xbps, and then people consistently get their maximum of 300xbps. But if you really want to push the limits on what's possible, a 400xbps rate may be possible sometimes but not at other times, depending on real factors of network congestion.

Re:VPS

[Anonymous Coward]

I used to do similar tests.

Either from a VPS, or shell account, perform tests with the following conditions. Avoid Amazon as some ISPs may assume it's netflix and do traffic shaping.
1. Via SSH, HTTP, and FTP.
2. Test file should be large, compressed, random binary file (so no compression optimizations done). So any compression overhead done by clients will yield zero advantage.
3. perform tests at same period of day if done multiple times.
4. plot results and measure averages.
5. Compare results to those from speedtest done at same time.
6. Consider using different size files, as some QoS systems will slow down after X amount of bytes transferred.

Re:VPS

[ledow]

If your port 80 is being throttled, it's being throttled. That's going to affect a random website as much as your VPS.

All we avoid is ISP's "unthrottling" select websites to give you a false impression that throttling isn't enabled. That's exactly what a VPS download will discover - the real download speed of your connection to a random website.

Re:VPS

[rjstanford]

If your disk is slower than your network connection then either you have terrible disks, or your network connection is just peachy and you should stop worrying about it.

Re:VPS

[sexconker]

The problem with FTP is that it can be a PITA to make sure it is measure memory to memory speeds and not disk to disk speeds. Sometimes you can write files to /dev/null and read from /dev/zero (and then manually abort it since /dev/zero is endless).

If your downlink out paces your disk writes, it's time to upgrade.

Re:VPS

[swb]

I usually like to build the file with random data and then run it through gzip to make sure the data is in no way compressible.

Re:VPS

[nabsltd]

If your downlink out paces your disk writes, it's time to upgrade.

Or, if you have a good SSD and your download is faster than the disk, your network is fast enough...stop worrying.

Re:VPS

[Nikker]

Even better idea:

Go to Low End Box and browse through hosts. Each one in the comments section will have a 10/100MB test file from one of their servers. As a bonus it will also tell you the geographic location of each server.

Re:VPS

[DrugCheese]

I was just using this method last night. Using aria2 and putting 4 to 8 connections through at once to see how fast I was going. It was in the ballpark of what I was expecting so I can see myself using it again in the future for ballpark numbers.

Re:VPS

[AK Marc]

That's wrong. When you have a read/write download, like a torrent, I've topped out a 5400 laptop at about 50 MBps. Maybe it was other factors, but the same file on a "better" computer did much better.

UC Berkeley's NetAlyzr.

[Anonymous Coward]

UC Berkeley's NetAlyzr.

ndt

[Anonymous Coward]

NDT - Argonne National Laboratory
ndt.anl.gov/

Not associated with any ISP.

There are other ndt (network diagnostic tests) as well.

Very detailed reports.

Re:ndt

[linear a]

Won't work if it's widely known. Speed test sites don't need to be in collusion. ISP's just prioritize their traffic. It's quite obvious with my ISP if I do speed test sites versus just finding something large to download from a cloud storage service.

Re:ndt

[jeffmeden]

Won't work if it's widely known.

Speed test sites don't need to be in collusion. ISP's just prioritize their traffic. It's quite obvious with my ISP if I do speed test sites versus just finding something large to download from a cloud storage service.

The obvious issue with that thesis is that you can't prove that the cloud storage site itself is performing slowly due to a bottleneck where it peers with your provider (or many other possible reasons) and while some providers are generally better than others about managing internal bandwidth, none can be said to have ALL uncongested peering points to ALL local customers and this obviously will have the same negative impact on user experience as a locally congested network.

Re:ndt

[jeffmeden]

Won't work if it's widely known.

Speed test sites don't need to be in collusion. ISP's just prioritize their traffic. It's quite obvious with my ISP if I do speed test sites versus just finding something large to download from a cloud storage service.

Doublepost...

One does not simply measure bandwidth

Without starting five or six torrents and leaving u/l and d/l limits turned off

Re:ndt

[Anonymous Coward]

I've done it. I've downloaded a large file from a friend that was taking forever, i.e. in the realm of two hours total time. A half hour in or so, I got sick of it and we both had bandwidth to spare. On a whim, he moved it to a folder on the exact same server called /speedtest/ and I tried again. The second download finished before the first one did and exceeded the bandwidth I was provided (my service is 2 MB/s and the file downloaded at over 3). Literally everything was the same, but one file downloaded at 3MB/s and finished in under 10 minutes and the other chugged along around 200KB/s.

You should try it yourself sometime. Just follow the speedtest.net directory structure.

Re:ndt

[linear a]

That's why you test multiple sources and see what the plateau of performance tends to be. You get various low values but will get a sort of upper limit where some values cluster. Except for, when i tested this, the widely known speed test sites I used were 2-4 times faster than anything else. This was a really obvious effect a year ago when I had satellite internet.

Re:ndt

[TemporalBeing]

Won't work if it's widely known.

Speed test sites don't need to be in collusion. ISP's just prioritize their traffic. It's quite obvious with my ISP if I do speed test sites versus just finding something large to download from a cloud storage service.

The obvious issue with that thesis is that you can't prove that the cloud storage site itself is performing slowly due to a bottleneck where it peers with your provider (or many other possible reasons) and while some providers are generally better than others about managing internal bandwidth, none can be said to have ALL uncongested peering points to ALL local customers and this obviously will have the same negative impact on user experience as a locally congested network.

I've actually used the SpeedTest sites to help improve downloading of Linux DVD ISO images. When I started the download (FTP/HTTP download) the quoted time was well over 8 hours, and the transfer rate was abysmal (60KBps to 120KBps on a multi-MBps line). Out of curiosity I ran a speed test through DSLReports and then found that the download rate jumped to 300KBps. After a while it would drop back to down to the previous range; I'd run the speed test again and voila, but up it went. I ended up downloading the entire Linux DVD ISO in under 1hour.

FYI, that was on AT&T DSL - not uVerse, just plain DSL since that is all we can get in our apartment. So obviously the ISPs are padding the numbers; which is a natural outcome of the FCC wanting people to report the ISPs that are not holding up.

Re:ndt

[complete loony]

And then set a lower u/l limit and watch your d/l speed increase. If your upload is congested, you won't be able to send TCP ACK's or protocol requests in a timely manner.

Re:ndt

[penguinoid]

Even if the speed test website is not associated with the ISP, there is nothing to stop the ISP from prioritizing the request because the ISP knows it's a test.

My recommendation: make your own test. For latency, send pings to various websites, disregarding any that are slow (since those are likely the website's fault). For bandwidth, use utorrent to download a large file with lots of seeds (such as a Linux iso).

Re:ndt

[the agent man]

not much to conclude with 2 data points.

Re:ndt

[meerling]

Sorry I can't remember it, but a few years back I used a method that obfuscates the site you are going to so the ISP doesn't "optimize" the speed test.
With the well known test sites, clear speeds vs obfuscated speeds were widely different. (Obfuscated was much lower.)
To try and make sure the obfuscation wasn't causing the issue, I also checked with several, at that time, barely known speed test sites as well. Of course, both clear addresses, and obfuscated addresses again. With the little known sites, both clear and obfuscated results were the same as the major sites obfuscated results.
That made it pretty clear that that the ISP was juicing the speed test sites when they could recognize it as one.

Re:ndt

[wzyboy]

Is that possible that your ISP has a web proxy set up between you and the public Internet? Some ISP will do that to save bandwidth. This may explain why the first test is slow and the second test is fast.

Re:ndt

[Slashdot Parent]

Not the OP, but presumably the second download had a different URL, so I'm not sure that any web proxy would have known to cache the file because it was the same as the first one.

Re:ndt

[Bengie]

I don't know if there is an official number or formula, but when doing iperf LAN speed tests with a 1500mtu, I'm seeing about a 30:1 bandwidth:ack ratio. So if you want 30mb down, you will need at least 1mb of free bandwidth up just for acks. That was using Windows and I know windows does support Naggle, which can combine acks to reduce their numbers. A naive TCP stack may need a 15:1.

Re:ndt

[almondo]

Interesting, traffic shaping based on URL path, cute, bogus, not surprising.

Re:ndt

[MPBoulton]

Interesting. Has anyone reported this to the FCC / your country's equivalent?

Speed test detection?

[Anonymous Coward]

They possibly have a speed test detection. It can be done by looking after "speed" in the url or with a list of know speed test sites.
They increase the speed of your line as long as the speed test is running.

You could work around that by "running" your own speed test in the background, but do it with a very low rate limit.
Could be done with something like "wget --limit-rate=x "

Re:Speed test detection?

[nazsco]

1. get a speedtest.something domain
2. patch a bittorrent client to register all your peers to that domain dynamically and connect via http to each other
3. ????
4. 1gb/s torrents!

DIY test

[Anonymous Coward]

If your ISP doesn't fiddle with your traffic, a heavily seeded torrent will normally do a good job of saturating your connection.

Re:DIY test

[ihtoit]

mod parent up. I find a newish torrent with a dozen or so seeders will saturate my connection (7MBit cellular).

Re:DIY test

[Bengie]

Saturating my download is simple, it's my upload that's hard. Few peers from the USA can handle my 50mb/s upload. When seeding Linux ISOs, most peers that max my connection are from Sweden or Germany and peer directly with Level 3.

Re: DIY test

[Anonymous Coward]

Torrent on cellular? You are why we can't have nice things.

Re:DIY test

[mythosaz]

Friends with internet.

Get one.

Test your upload. It will have the simultaneous benefit of testing their download speeds.

Re:DIY test

[Bengie]

I was able to get an early torrent of a big Linux distro, like Ubuntu right at a major version release, and I had hourly averages, according to PFSense, of 42mb/s-46mb/s for almost 8 hours strait from 3pm-11pm. That's an average of ~90% utilization. I was limiting Torrent because I didn't want it to interfere with my video games.

Re:DIY test

[nabsltd]

Saturating my download is simple, it's my upload that's hard.

You could also try for total composite speeds by using a bunch of torrents. My seedbox regularly runs 40Mbps total upload speed on the 50+ torrents, even though no one torrent is running that fast. I actually throttle the max upload to 50Mbps to allow 30Mbps free bandwidth to FTP the files to my home.

Re: DIY test

[ihtoit]

yes, I'm the cunt who takes what he's paid for and is thus entitled to - unlimited data plan.

Re: DIY test

[Barny]

If they advertise it as unlimited, it should be unlimited. Download more please!

Re: DIY test

[ihtoit]

thank you, I do believe I will! :D

kernel.org? ftp.insert-ftp-site-here.whatever?

[davidwr]

Seriously, find a handful of known-high-bandwidth places to download stuff from and download some large files from each of them and use your PC's network-monitoring tools to gauge your bandwidth.

As for as upstream, get some email account from various providers, compose a message, and attach a large-ish file.

Note - if your ISP gives you "burst speed" you will have to "burn through that" before you start getting "real" numbers.

Re:kernel.org? ftp.insert-ftp-site-here.whatever?

[Barny]

Hint, nvidia drivers are a good one to use. They use akami I believe, and you should be able to max out damn near any connection with a few files coming from there.

I use speedof.me

[AvitarX]

And frequently score higher on my tmobile phone than on comcast (up to 30 vs up to 15)

Re: I use speedof.me

[AvitarX]

Im trying to compare apples to apples though ,so I do 4g vs WiFi on phone .

Re:Ask yourselves these questions... apk

[ColdWetDog]

All right. Just who woke him up this time? Whoever it is, you need to put him back down in his bunker and this time LOCK THE DOOR.

Re:Speakeasy Speed Test

[leonbev]

That appears to be run by OOKLA, the same guys who run Speedtest.net.

I don't trust Speedtest.net's results, either, as they seem to ALWAYS run at the maximum speed for the connection even when my Internet connection on sites like Youtube or Netflix is slow. I think that there is some shady content prioritization going on there.

SamKnows from the FCC

[NevarMore]

https://www.samknows.com/

I have one of their boxes installed. It seems to provide a clear picture of overall performance with a monthly report. I'm doing this because I'd like to think it helps the FCC keep the ISPs honest.

PS - Card carrying Libertarian. No the FCC isn't spying on me, and yes regulation of ISPs is appropriate. If we've broken the free market by granting a local monopoly or limited oligopoly then heavy regulation is appropriate. Consumer choice is better, but this is the best we can do with what we have today.

Re:SamKnows from the FCC

[Anonymous Coward]

I too am a samknows house and I can confirm that my DSL speeds, uptime and whatnot have been absolute and consistent since the device was put in service over a year ago. I know the ISP, mine is Centurylink, is also notified that you're a samknows location and I suspect they make sure you get what you pay for. Could be a coincidence that my DSL has been running perfect since I put this in place but I doubt it.

Re:SamKnows from the FCC

[Nkwe]

https://www.samknows.com/

I'm doing this because I'd like to think it helps the FCC keep the ISPs honest.

It probably also helps to ensure that *your* connection gets priority...

I have one of the boxes as well and ensuring that my ISP is motivated to give me good service was part of the reason I put it in. I also think that it is a good idea to have a FCC based performance monitoring infrastructure out there. While I don't think the program is monitoring user's activity, I am a bit on the paranoid side, so I don't run all my traffic through the box (which is a supported configuration.)

As an aside, I am on FIOS and according to the FCC monitor box, I almost always get my advertised speed. My gut feeling with day to day operation of my connection is that the FCC box is giving me real numbers.

Re:SamKnows from the FCC

[Anonymous Coward]

Thank you for mentioning that. I am a FIOS customer in the Washington, D.C. area. I regularly interact with remote machines at my employer in North Carolina. I have the Verizon FIOS 25/15 plan. During normal business hours, it works great. But starting in the late afternoon, usually around 5-6 pm every night, round trip times go to crap. My tracert experiments show that it is almost always an ALTER.NET (a.k.a. Verizon) to level3 hop that introduces the latency.

I raised this on a Level 3 blog (Mark Taylor, article titled "Verizon’s Accidental Mea Culpa") and was told that this is another case of Verizon refusing to add ports to its interconnect to Level 3. I presume this is Verizon's way of neglecting to provide peering that is optimal for Netflix customers until Netflix either pays it or the net neutrality regulations get resolved. Meanwhile, I'm caught in the crossfire trying to do work over a basic remote desktop connection, which sucks when latency is high and/or packets are dropped. It is not the bandwidth that is the issue here--I'm not doing heavy downloading at all--it is the latency and/or dropped packets.

Have you experienced this kind of problem, and did it change after you installed your samknows box? Thanks...

Re:SamKnows from the FCC

[Archangel Michael]

Other Libertarian here. ISPs are a product of government collusion (monopoly) practices. The further a government is away from me, the less I trust it.

That being said, we do not need MORE regulation of ISPs, we need out of the box thinking to reframe the last mile problem. I have NO problem with a local municipality running last mile service, as long as I get to choose which provider gets to my house. This would require Fiber to the premise, running back to a COLO that is rented (funded) out to service providers to provide ANYTHING they want to the customer. A choice of four or six providers at the COLO to choose from, and a simple Network VLAN change to change providers.

THIS would negate the need for ANY regulation. If Johnny Christian wants to have Jesus.net cable co, he can get it. If Mary Rotten wants all porn and drug channels, she is free to choose that. Comcast will be forced to stop playing games, because they will lose customers if they throttle YouTube and Netflix.

Almost all problems we have right now, are caused by last mile monopolies. Lets inject CHOICE, rather than regulation into the market.

Re:SamKnows from the FCC

[Nkwe]

Thank you for mentioning that. I am a FIOS customer in the Washington, D.C. area. I regularly interact with remote machines at my employer in North Carolina. I have the Verizon FIOS 25/15 plan. During normal business hours, it works great. But starting in the late afternoon, usually around 5-6 pm every night, round trip times go to crap.

[...]

Have you experienced this kind of problem, and did it change after you installed your samknows box? Thanks...

I am in the Portland, Oregon area. When FIOS first came to the area it was put in by Verizon. Since then they have sold the assets to Frontier, who now runs my FIOS. My experience has been that I have always gotten full speed from my FIOS connection under both Verizon and Frontier and both before and after the SamKnows box was installed. I haven't seen a change in behavior. Note that I have had my FIOS service since 2007 and participated in SamKnows since 2010. The public NetFlix related complaints with Verizon FIOS are relatively recent, so I don't know if my continuing good performance is due to me now being on Frontier, the SamKnows box, or just good luck. I am also on a business FIOS connection (to get static IPs), so that may help as well. I have had excellent customer service with both Verizon and Frontier over the life of my FIOS connection.

Re:SamKnows from the FCC

[Anonymous Coward]

This would require Fiber to the premise, running back to a COLO that is rented (funded) out to service providers to provide ANYTHING they want to the customer. A choice of four or six providers at the COLO to choose from, and a simple Network VLAN change to change providers.

That's what they already have in Iceland, Sweden, Finland, Norway, Denmark, and the Netherlands!

Libertarians dream of a world with X. Socialists live in that world.

Re:SamKnows from the FCC

[gweilo8888]

You should probably read this: http://www.farces.com/samknows... Samknows is frankly a waste of your power bill; it doesn't keep ISPs honest in the least.

Re:SamKnows from the FCC

[gewalker]

According to this article, Johnny Christian and Mary rotten may both want to use the same ISP.

Re:SamKnows from the FCC

[Archangel Michael]

Because small monoculture Scandinavian countries are the same as the Melting pot of the world that has a multicultural world view.

Re:SamKnows from the FCC

[Archangel Michael]

Just to be sure we're talking about the same "choice", I see that Sweden has two or three Cable providers. The idea of a Cable provider going to COLO facility is crazy, unless they are splitting available Cable frequencies. To be honest, it looks like the "choice" is pathetic even if it does what I suggested. Then I look at the number of Cable Channels offered and I think I currently have more shopping channels those cable providers carry (slightly sarcastic). the number of channels is pathetic.

Which is, exactly what I would expect from Socialistic monopolies.

No, Sweden doesn't do what I suggested. Not even close, in practice or in principle. I suspect that my idea of "choice" is really alien to Scandinavians, who are used to their two (or three) crappy choices and think they're awesome.

Finland MIGHT be what I expect, however, I suspect that there are just a number of small city/regional cable companies.

And from this list, "No" the countries you listed do not do what I suggested. Not even close. http://en.wikipedia.org/wiki/L...

Re:SamKnows from the FCC

[Archangel Michael]

Possibly. Though I don't think liberals want to watch much of Joel Osteen. ;)

Re:SamKnows from the FCC

[stardaemon]

Cable? Wouldn't know about that market, since I don't watch TV.

But I do have FTTH and eight different ISP:s to chose from.
My current fiber options:
http://skb.qmarket.se/privat/b...
I live in stockholm, btw.

Google claims that "1 US Dollar equals 7.22 Swedish Krona", if you wish to do a price comparison.

There are plenty more ISP:s if you prefer ADSL. Don't know why you're so hung up on cable. There are plenty of other options.

Re:SamKnows from the FCC

[Archangel Michael]

Thank you, but I think that makes my case about last mile being the problem, not regulation. Competition solves many problems.

Re:SamKnows from the FCC

[mgcarley]

These are not the only countries to have open last miles.

Bit Torrent

[bigmo]

It's hard to know if slow speeds are from your connection or the server you're connecting to or something in between. If you download a linux distro over bit torrent you'll be bypassing any individual server bottleneck and any (except local) general network slow downs. I usually get extremely good speeds from bit torrent, pushing 15 mbit, from my "15 mbit" fios connection. I don't use it a lot so I don't see any alleged throttling from it.

DSLReports or any of that stuff is only useful to determine if you have a decent working internet connection. They should never be used for any sort of benchmarking as one has to assume carriers optimize connections to them to make themselves look good.

Re:Bit Torrent

[Teun]

The well respected guru's of my ISP agree with the other guru's that frequent the relevant newsgroup the only way to establish your line speed is to up and download from the ISP's own servers, using anything else is unreliable as it involves multiple parties that are not part of the contract you have with your ISP.

So they made available a bunch of files to play with.

ftp://ftp.xs4all.nl/pub/test/

Re:Speakeasy Speed Test

[Anonymous Coward]

T1 - is it 1999? Where is Prince...

you don't understand

[Anonymous Coward]

you need a comparable site to where you are, due to the nature of the internet.

the best way to test this would probably be to pull a file from the common CDNs, so Akamai, Amazon CloudFront, CloudFlare etc...

Speeds depend on:
* The speed of your local connection.
* The contention in your ISP networks.
* The contention on your inter ISP links, and their peering arrangements.
* Speed to Netflix, YouTube, Amazon Video, etc..
* ....... etc....

So considering much of your traffic is to the closest CDN, and video streaming nodes, that's what you really care about.
If everybody tried to stream a file from Australia, only the Australians would get good results as the rest of the world would experience TCP window size congestion slowing their transfers and we'd end up with a map of the latency from the Aus node rather than broadband speed.

Apps

[NuAngel]

I use the App for Windows 8 and Windows Phone 8:
Windows: http://apps.microsoft.com/wind...
Phone: http://www.windowsphone.com/en...

Re:SpeedTest.Net

[Bengie]

Speedtest varies a lot depending on which server you select. Some server have a nice long test that tends to return proper values, but some are so short, it can't possibly be accurate. I have had quite a few times where Speedtest would say something like 70mb/s on my 50mb connection, but when I look at my local computer's bandwidth usage and my PFSense firewall, I'll see something like 42mb/s.

I don't really think a 1ms away ISP speedtest counts much either. Show me full speed from a server 200ms+ away. I can get my full 50mb/s from German servers that are 210ms away from my Midwest USA location.

Youtube speed test won't tell you anything

[DiSKiLLeR]

Youtube speed test won't tell you anything as youtube content tends to be cached locally at your ISP by GGC (Google Global Cache).

Re:Youtube speed test won't tell you anything

[PRMan]

And yet, for about a year with Time Warner, YouTube was the ONLY bad site. Constant and endless buffering. Until I set my preferred video to 720p. And then it ran much faster than the 240p, 360p or 1080p.

Then someone told me this could be fixed by switching to Google's DNS and sure enough it went away immediately.

Re:Speakeasy Speed Test

[Anonymous Coward]

That's because your ISP throttles Youtube and Netflix.

No

[Charliemopps]

There is no way to test "The internet"
The fact of the matter is you make dozens of hops, even hundreds, to get anywhere. En-route you can hit any number of choke points. If you run a speed test I can almost guarantee your ISP knows about the speed test site and is going to prioritize your traffic. Add to that the fact that the speed test site is likely hosted somewhere like the Amazon cloud and all you're testing is your route to about the easiest place to get to.

Is your ISP throttling Torrents? Netflix? Youtube? A test to any other site is useless if they prioritize that and throttle where you actually want to go. Is there a problem with your NID? The remote you connect to? The peering they have setup?

On top of all of that, speed test sites are just a test of downloading various file sizes. That's easy... flawless movie playback and seamless online game play? That's an entirely different story. You've no idea how many friends I've had complain about their ISP throttling their game, only to find out later the problem cleared up when they got a new video card. lol

So if your ISP is not working for your needs, you need to switch. If you have other options, most offer a contract free option now-a-days. Try that out and cancel if it's no better. If you have no other options, you're stuck with it anyway.

Your best bet, if you're stuck with that ISP, is to make friends with a tech. Get one out there for some reason, offer him a beer, whatever. Joke, laugh, etc... he'll probably tell you what's up. Once you know where the problem is, often you can figure out how to talk them into a better solution. In these situations you're usually fighting their bureaucracy... its not that they don't want to help, it's just a lot of paperwork to get that help. Be more annoying than the paperwork.

Re:No

[thegarbz]

If you run a speed test I can almost guarantee your ISP knows about the speed test site and is going to prioritize your traffic. Add to that the fact that the speed test site is likely hosted somewhere like the Amazon cloud and all you're testing is your route to about the easiest place to get to.

I fail to see the problem with this. If the goal is to test my line speed to confirm that there's nothing wrong in my house or in the street connection then this would be a perfect test.

Wanting to test anything else is just ignorant as you mentioned due to the number of possible choke points. But then really why not just look at your download speed. Why would you care about doing some "real world" testing if you don't care about your "real" download speed at any given time?

Re:No

[Charliemopps]

Let me explain...
Lets say you live in a suburb outside of Dallas TX
Your ISP is based in Dallas
Your ISP has a 10gig tunk from the suburb to Dallas
Your ISP then has 100gig worth of peering to the rest of the net from Dallas
You notice that when you try and watch Youtube, you're having trouble
On friday night, when you decide to test, they have 130gig worth of traffic coming out of your suburb. (i.e. that trunk is over-saturated)

If traffic to your testing site is prioritized, and I guarantee you that it is, any test you make to that site will work fine.
Everyone else, going everywhere else on the net will be forced to slow down a tad so your test can have an unimpeded route.
Viola, you do have a saturated trunk causing you problems but the test does not reveal it.
The only way the test would actually work is if everyone on that trunk tested to the same site at the same time.

Given modern traffic shaping tech, they can shape your traffic based on almost anything. Port, site, packet type, anything. It's virtually impossible to "test" anything on the network unless you own the network.

Could the test reveal a problem? Yes. But if the test shows there is not a problem, it definitely does not mean there isn't one.

Re:No

[thegarbz]

You notice that when you try and watch Youtube, you're having trouble ... snip ...Viola, you do have a saturated trunk causing you problems but the test does not reveal it.

I've highlighted the important parts of your comment. Youtube is having trouble. I don't need a test to show me that. The test shows me I'm getting full speed. Done end of story. The problem I am having is not my problem it's my ISPs problem. I don't need any further information beyond that specifically because there is absolutely nothing I can do about this.

If the speed test was showing crap, or I was downloading a file from my ISP at crap speeds then I would call the ISP and say I have a problem. But the peering arrangements are out of my control, and I don't need a test to establish this as the reason I would be looking it up is if I notice a problem to begin with.

Speed testing the entire internet is a pointless exercise. Speed testing your last mile is important.

Re:Speakeasy Speed Test

[Nethemas the Great]

Network topology isn't that straight forward. From your ISP's routing center to different portions of the Internet can be faster or slower than others. Check out network peering topic to understand why YouTube may be slow, while other sites are not.

Xfinity Speed Test

[jbov]

Use the Xfinity speed test at speedtest.comcast.net.
As far as I can tell, they are not affiliated with any ISP.

Re:Xfinity Speed Test

[SemperUbi]

Either that or some impressive cognitive dissonance.

A few options...

[Twinbee]

Speedtest.net used to be good at one stage. But when I tried them relatively recently, I found that they measure the speed once it gets going, and ignore the regular dropouts that may occur. Speedtest.net claimed about 1gigabit, but in reality it was a tenth or even a fiftieth of that.

I had more luck with the following:

http://speedof.me/ - HTML5 Internet speed test (no Flash or Java needed). It claims to be the "smartest and most accurate online bandwidth test".
http://testmy.net - Nice graph and intelligent picking of the size of the test file to download.

Short answer: No

[Morgor]

I'm a network engineer at an ISP, so I would say I have a bit of experience with this from both ends of the table. First of all, there's a difference between your broadband connection speed and your perceived rate. Your broadband connection might be capped to what you pay for, and, assuming your last-mile medium can handle that speed, that only means that you will never actually go beyond your connection speed.

Now as we know, the internet is a complicated network of interconnected systems. You are connected via your ISP's backbone to the other systems (ISPs, enterprises, content providers, etc.) via a number of internet peering points. These peering points have their own connection speed (typically 1 Gbit/s or 10 Gbit/s, although higher exist), and may or may not be utilised to their maximum extent at any point of time. This means that you may have your full data rate available to some destinations, while others may take a congested route.

You mention testing, and your frustration is very reasonable. There are testing sites out there, but you never have any idea about how many else might be testing at the same time, or how much load there is on the server at the moment of the test. If you are unlucky, you might also be limited by your hardware, your operating system (TCP Window Size, receive buffers and similar might not be tuned properly), or your router.

I would say your best choice would be to download as much as possible from as many sources as possible (bittorrent is excellent for this, but may be throttled by evil ISPs), and do this over a couple of days to get an average indication of how much your connection is capable of delivering.

If you have a server on some remote location via the internet, you can use programs like iperf to make a bandwidth test, but such a test is not exactly precise when you have no idea how the intermediate networks are.

Re:Short answer: No

[Morgor]

Well, you have to understand that traceroutes are replied by routers, and routers might choose to prioritise ICMP (the protocol used by traceroute replies and ping) lower than normal traffic. So you should only use traceroutes to get an idea of the path the data takes, not the health of that path.

Re:Short answer: No

[lexman098]

I think part of the responsibility of the ISP to make the correct peering agreements such that my traffic isn't appreciably slowed within a reasonable area. When I use the default speedtest.net settings I usually see perfect speeds, but that website lets you select further out locations and often the next state over gives me horrible speeds. This is corroborated by lag in video games etc. The point here is that it's not just the ISP's responsibility to give me the advertised speed within their network. Those speeds are irrelevant. I want decent speeds all over my section of the country (at least), and it certainly *IS* the ISP's fault from my perspective if they can't properly deal with their peers.

A few options.

[B5_geek]

If you have a system that you can test against (i.e. a server at your work with a fatter-pipe then you have at home, or a hosted server/VPS/etc.)

iperf

run "iperf -s" on the server and "iperf -c server.ip.address" on the client.
Read the man pages for more options.

If you don't have a 'known better then you' to test against try this to test your maximum download bandwidth.
Simple test: download a large file from Microsoft (i.e. a 'network install' service pack, or similar) or other big-host

More complicated:
run several (4-20) 'wget' concurrently. If you use Linux .iso's as your target download, make sure you grab the files from *.edu sites. Schools should have a lot more bandwidth then the average .com that is hosting files.

Your ISP might have several things in place from preventing DDOS attacks from there customer machines. So each 'download' might be throttled by your ISP. If you open several download threads to different locations, downloading different things you can maximize your usage.

Also, don't download the same thing twice from the same source. Caching can/will interfere with accurate measurements.

Re:Speakeasy Speed Test

[NoNonAlphaCharsHere]

return ArtistFormerlyKnownAs(ArtistFormerlyKnownAs(Prince));

Google & ISC have MeasurementLab.NET

[Hobart]

The Network Diagnostic Test was able to see performance problems on my cablemodem connection that Ookla's speedtests did not.

http://www.measurementlab.net/...

Unfortunately, the number of ridiculous hoops you need to go through to let an unsigned Java applet run an arbitrary network I/O makes it much less useful.

Re:Google & ISC have MeasurementLab.NET

[nabsltd]

http://www.measurementlab.net/...

Unfortunately, the number of ridiculous hoops you need to go through to let an unsigned Java applet run an arbitrary network I/O makes it much less useful.

They now have a Flash version as well, so it's easier. But the numbers appear really low, claiming that my network buffer limits download to 140Mbps, yet I have often downloaded actual files from the Internet at faster than that.

OTOH, all the Ookla-powered sites claim I get over 70% of my 1Gbit network card speed, which I also find hard to believe, despite having a 20Gbps connection to our ISP (with literally thousands of users, one of which is a server I maintain that downloads at over 2000Mbps 24/7 backing up a remote site).

functional test

[stanglover]

I always prefer a functional test.... Download steam then a f2p game. Check your download rate. Delete the game then retry it on a cable connection.

BitTorrent or some other p2p file downloader

[mi]

Pick a popular torrent — like a recent release of your favorite BSD or Linux distro — and start downloading (without any limits on your client side, of course). Watch the bandwidth. With a large number of peers, your measurement will be insulated from the oddities of any particular connection.

Re:BitTorrent or some other p2p file downloader

[Teun]

Or download an Obama interview.

Re:BitTorrent or some other p2p file downloader

[afidel]

If you have a typical asymmetrical connection you'll want to limit your number of peers to ~200 per 1Mbps of upload you have, any more than that and you tend to actually see your download speeds slow as your client uses so much bandwidth managing peers that it chokes off the return packets to keep the download speed going.

To many variables

[BringsApples]

There are simply to many variables to get results that you feel comfortable believing. But what you could do is create a file that's 10M and send it and receive it from a friend (or multiple friends) connection that is using the same bandwidth speeds and a different ISP. Even this has lots of holes in it, but at least you can get some peace of mind. In my mind, there is no true valid test, unless you have complete control over all hardware between you and the end, which not many can have.

Another way to get peace of mind is to just use all of the sites that measure bandwidth. If they all report basically the same thing, then take it as fact.

Two cents, two alternatives.

[TechkNighT_1337]

I know it's not a good reliable test, but you can always try do download an .ISO file from some Linux distro from various sources or some big program from sourceforge. The second alternative, you can try to use the meter from the Brazilian agency for internet at: http://simet.nic.br/medidor/ (try googling: simet nic br) it's not in any form affiliated with any US ISP and i think we have sufficient bandwitdth for the test.

Usenet

[RobbieCrash]

Download some binaries from a Usenet provider, that'll max out your connection.

I generally get ~13.5MBps down on my 120Mbps connection from Rogers. Uploading to my VPS gets me a solid 2MBps out of 20Mbps.

Re:Usenet

[Spritzer]

That's my method. 10 connections per server to 3 servers significantly geographically separated.

Grown Up Terminology

[rjstanford]

Really?

Either just say, "... my ISP, Comcast..." or don't name them at all. Trying to be cute just muddles the conversation and gains absolutely nothing.

Why do you care about other people's results, too? Just upload a large file to somewhere with known good bandwidth (amazon S3 might be a good choice, or FTP it to Dreamhost, or whatever), time it, then pull it back down again (and time that). You'll get a pretty accurate "actual bandwidth" there.

If you're paranoid - and it appears that you are - make the file something unique and check the checksums in both places (or just record a brand new 60 second video, timed upload it from one machine, then timed download it to another and play it). No way that anyone can optimize that transfer - if they could, they wouldn't be wasting the technology on you (and, quite frankly, if they could move 7 megabits/second over a "5 megabit/second pipe" then they'd be entitled to say that they had a 7mpbs pipe.

Not everything needs a dedicated app.

Re:SpeedTest.Net

[rjstanford]

They all count... on the one hand you're testing your local loop, on the other you're testing your ISPs peering. Both are valid tests.

Shaper Probe

[QuietLagoon]

Not run by ISPs. Sniffs out bandwidth shaping.

.
http://www.measurementlab.net/...

Runs on OS-X, Windows, Linux. Port available on FreeBSD.

Stream test urls

[Nonesuch]

Netflix offers several test streams for validating your speeds, and Google has a Video Quality Report

I find that the Speedtest.Net results are a realistic estimate of my actual best case upload/download speed, but there are certainly some websites which are much slower to load, for various reasons. If you suspect your ISP is throttling some websites intentionally, you can always browse through a VPN service.

As mentioned previously, local WiFi problems are often the root cause of slow page loads. Go wired. You can also use the network debugging tools built into Firefox (Network Monitor) and MSIE to try to determine what parts of a page are particularly slow.

Re:Ask yourselves these questions... apk

[rjstanford]

Even being noisily right with an answer to a question that nobody's asking, in a conversation about something completely different, is annoying and should be discourages.

Re: Speakeasy Speed Test

[Phreakiture]

Change ISPs.

To what?

Re:SpeedTest.Net

[QuietLagoon]

Be careful with Speedtest. They use repeating ASCII data, which compresses very well. This can lead to misleading results in some instances.

Re:Speakeasy Speed Test

[Phreakiture]

I thought this at first, also, but I have had a pretty close match to Speedtest's claims when using scp to send large files to/from my EC2 instances.

Re:Ask yourselves these questions... apk

[Anonymous Coward]

You can't prove a flat-earther wrong either because they will ignore or dismiss any amount of evidence. You could take them up into space and show them the spheroid Earth for themselves and they would say it was a trick or illusion. Arguing with APK and his faggot fanboys is the same. You will always fail if "success" means "convince APK and his ilk". You will always succeed if "success" is "give him enough rope to hang himself and let everyone see for themselves what an obsessive, single-minded, raving lunatic with no life he really is".

None of them

[xushi]

Stick to old fashioned measures...

Just get a copy of httpwatch/fiddler, (wireshark too if you know what you're looking at), and test several hits with a few random sites. Preferably the big sites that tend to have more capacity (news sites, apple, etc..).

Not only will you get the big picture of how long it took to load, but you can also focus on what took too long to load - if it's a network or site issue, name resolution, etc..

ICMP is also your friend (latency test).

Try popular torrents

[mike449]

Downloading a popular TV show episode over Bittorrent will saturate your link. It is a good measure of your connection speed.

Android FCC Speed test

[hAckz0r]

The FCC has an Android App that will test your data connections from your phone, and allows them to monitor your provider bandwidths. When your phone connects to your local WiFi the app is testing your Cable ISP, and when not, its testing your cellular ISP. In both cases the data is collected by the FCC to make sure your bandwidth is not being throttled unnecessarilly.

In theory the ISP's might look to see where your data is headed and make adjustments based on that, but that of course would be deceitful. No, they wouldn't do that would they?

VPN improves my net performance and test results

[Optic7]

I had been noticing poor performance from Youtube when watching videos (buffering, dropping to low-res, etc). Then I noticed that youtube seemed to work much better while I was connected through VPN, which is the opposite of what you would expect, at least in theory. But I realize that ISPs have been playing throttling games with large video sites like Youtube and Netflix.

However, I did another test and the results of it were more surprising for me. I have 3mbps DSL service through Verizon. If I run a test through speedtest.net, it reports right around 3mbps. However, if I connect my VPN first and then do the same test, it reports around 5mbps! How is that even possible?

Unfortunately, I feel like the VPN slows normal browsing of other sites a little bit, but I haven't done a comparison yet to confirm my perception.

Re:Speakeasy Speed Test

[lgw]

Youtube gets plenty screwed up without ISP throttling. There are days when I can't watch some videos on Youtube at the lowest resolution, but others are fine in HD, and my ISP is on Google's "nice" list for Youtube.

Re: Speakeasy Speed Test

[Dishevel]

T1, while slow, has its place.

1.5 Mbs, synchronous, rock solid and dedicated.

If you can afford it. You do not need faster and it has to be up it is great. Also I have 4xBonded T1 as a backup to my 50 Mbps fiber at work. The copper and fiber runs are completely different and gives me awesome piece of mind.

Test peering

[imcdona]

I'd like to see a site that's capable of testing the peering throughput of a given ISP. As opposed to running a single speed test, it'd run run multiple tests consecutively via servers on different networks. Take the Verizon fiasco recently where they had a saturated link to Level 3 that affected Netflix. A peering test would be capable of highlighting this sort of thing.

Re:VPN improves my net performance and test result

[apraetor]

Does your VPN use compression? Try running the VPN-routed test using testmy.net; they use random incompressible data for their testing.

Meaningless

[jtara]

These speed tests are basically meaningless. There are too many factors that might affect the throughput and latency from your desktop or device to any given site.

Meaningful tests might include:

- local link test to neighborhood node, Internet access point - your ISP would need to install test servers in local (neighborhood, at least for cable setups) nodes and wherever traffic exits their network to the Internet. This would allow you to test latency and throughput within your ISPs own system. Obviously, this ultimately limits possible Internet speeds. Your ISP almost certainly already has these kinds of test servers. But they may or may not expose them or advertise them to users.

- A test employing MULTIPLE SIMULTANEOUS test servers. This would at least attempt to assess your available bandwidth "to the Internet".

You should not have any reasonable expectation of achieving the maximum theoretical throughput of your "Internet connection" to any given site. Or any one site at all. I do not know why people obsess so over these meaningless tests.

They're not necessarily trying to trick you

[nine-times]

I don't trust the usual sites, the first ones I found via Google. I suspect (and found) that at least some of them are directly affiliated with ISPs, and I further suspect that traffic to those addresses is probably prioritized, so people will think they're getting a good deal.

I just wanted to point out that they're not necessarily trying to trick you by running these speed tests. For one thing, if they wanted to trick you, they could always just compile a list of popular test sites and prioritize/uncap that traffic.

But it's actually somewhat valid for ISPs to provide tests that, in a sense, are biased. Let's say you have a Verizon connection. Verizon may want to provide a testing mechanism to make sure you're getting the advertised connection to their network, to make sure things are operating properly. If you have a slow connection to Slashdot, for example, that might just mean that Slashdot is slow. It might mean that your route to Slashdot has been saturated somehow, and that might not be Verizon's fault. There are a lot of things that could possibly go wrong that could cause your connection to Slashdot to be bad, and Verizon can't rely on that as a good test.

So what Verizon would want to do is provide a test that simply confirms that your connection to their network is running at advertised speeds, which would mean testing between your home computer and another machine on their network. If that is operating at advertised speeds, but your connection to some endpoint is slow, then the problem is probably between Verizon's network and the endpoint, and not between you and Verizon's network.

Re:VPN improves my net performance and test result

[Optic7]

I don't know if my VPN uses compression, but that's a good tip. I will check that out, thanks!

Re:Speakeasy Speed Test

[SirCowMan]

A lot of the speedtest.net servers *are* ISP's. Many ISP's use their own servers through speedtest.net (or a dedicated page they host to the same servers) to verify installations, etc.

I've worked for an ISP

[Tyr07]

The problem you're going to run into isn't even 'bribed and biased' speedtests themselves.

The problem is they literally priortize the traffic the IDENTIFY as a speed test.

So you can have a completely neutral party and if they don't hide that it's a speed test, it'll get prioritized.

E.G Download file45456.zip from a server. Measure that speed, repeat to confirm.
Rename the file on the server speedtest.zip and bam, suddenly it gets way better speeds.

Re:I've worked for an ISP

[mysidia]

You just gave me an idea....

I'm thinking about making a VPN Service that "Looks like" a speed test.

Very simple.... you request a HTTP download of file45456.zip and a simultaneous HTTP upload of file45457.zip

To maintain the connection, your VPN client will do this repeatedly.

However.... inside the HTTPS transfer there will be the encrypted IP packets you are exchanging encapsulated.

Also... of course, the same website will have a speedtest, all over HTTPS :)

Re:I've worked for an ISP

[Tyr07]

If you come up with a way to spoof them into thinking regular traffic is a speed test it will be a beautiful thing, and you'll have to share it.

Re:How about more websurfing speed/bandwidth?

[mmell]

Take it easy on him, guys. He was abused as a child.

You still haven't sought out treatment, Alex?

[mmell]

You know, the abuse you suffered as a child will only continue to haunt you until you seek out competent mental health care. I don't know if it was your mother, your father or both that abused you, but they left damage that is clear for all to see. Please seek out psychiatric care immediately. Your constant claim to be trying to make the internet better by making it worse is clear evidence of your need for help. Seek it out before you destroy yourself utterly. I'm truly afraid that you'll actually commit suicide, like you pretended to do several years ago. You do remember posting in your mother's "voice" about how we horrible slashdotters had driven you to suicide, don't you? Please get help before that becomes more than a fantasy in your tortured mind.

Speed test for what ..

[savuporo]

Forget the speed test. Give me an accurate uptime / reliability test instead - i'll pay for it. In fact, i'll pay for a home router that has a service integrated for reliability monitoring.

I'll be happy to pay more for extra reliable service, rather than variably available peak bandwidth.

The only accurate test will be software-based

[mysidia]

Client-software based. The problem with any server-based test is twofold:

• There will be many variables in regards to the performance of a download/upload to some servers
• Web browsers are poorly instrumented. Most of the speedtest applications are using Flash or a Java applet. Java-based tests seem to often be more accurate. Flash-based speed tests seem to give inaccurate results, even for the performance of the server you are downloading from: it is very inconsistent

The ideal speedtest would look like a Knoppix CD... you'd boot your computer, and load up a Kernel patched to gather Web100 metrics, and then test a large number of upload and download scenarios against a large number of servers.

Then analyze the data to calculate a "Composite speed score"

Re: LMAO - Man: You have lost your mind... apk

[mmell]

I'm not the one posting anonymously, pretending to be several people and resorting to ad hominem attacks on any who point out everything you're doing wrong.

Please get help. I am not a licensed mental health care professional and even so I can plainly see evidence of your need for therapeutic aid.

Re: How about more websurfing speed/bandwidth?

[mmell]

The only thing I'm projecting onto you is the clear light of reason. Your mental incapacity (and its source) are both plainly evident for all here to see.

Speed Test Problems

[DERoss]

Most speed-test Web sites fail to tell the user where the the server at the other end is located or who owns it. For that reason, I generally use Speedtest.net or DSLReports, both of which allow me to select a distant server. Speedtest.net has a really large set of responding servers all over the world. DSLReports has a very limited set of servers for its Flash-based test but seems to match Speedtest.net for its Java-based test.

I have a browser extension that obfuscates my browser's outgoing HTTP headers and thus confounds many geolocation algorithms. Both Speedtest.net and DSLReports generally think I am someplace other than where I really am, in some cases on a different continent. I am not sure what is being tested in this situation, so I generally disable the extension.

"...get on topic"? that's rich!

[mmell]

Is your grasp on reality truly that weak?

No, that hasn't been the subject for a while now.

[mmell]

Nobody here is foolish enough to run a kernel-mode host file manager written by a man who clearly has what is at best a tenuous grip on reality. I know the truth hurts, but until you accept the truth and do something about it you will remain an object of derision and ridicule here.

Re:"...get on topic"? that's rich!

[gargleblast]

It probably is not. Deep inside, the narcissist is aware that his life is an artifact, a confabulated sham, a vulnerable cocoon.

Filesharing sites

[burning_plastic]

I've found the best speedtest solution is to choose a couple of the major fileshare sites (ie. the ones that ISPs love to throttle or block). I then upload a non-compressible junk file of about 100MB and try downloading it directly and via a VPN. I then compare those results to my ISP's speedtest.

I've seen some impressive throttling or within-ISP network congestion/lack of interconnects.

APK, you're showing your insanity.

[mmell]

Or do you think anybody here is anile enough to believe that this isn't just you?

Get some help . . . soon.

Is that the best you can do?

[mmell]

Your grip on reality is even weaker than I'd thought. I'm certain (based on your other posts) that you were abused by your father, and possibly your mother as well. Were you also sexually molested, or was it only physical and emotional abuse you suffered?

The linked text appears quite applicable.

[mmell]

Sadly, it probably is beyond my current comprehension. I am not a trained psychiatrist or psychologist.

Then again, my understanding is not what is required. If only Alexander Peter Kowalski were up to the simple task of honest self-appraisal, there might be some hope.

easiest way

[Anonymous Coward]

go to pirate bay, download a torrent with hundreds of seeders. wait 5 min and see how fast it's going. you now know your max download speed. hurray.

You are a sad little child.

[mmell]

Little wonder that you are your only supporter here, your only friend, your only fan. It must be lonely, living in a world all of your own. Don't you ever wonder what it would be like to enjoy friendship, respect, happiness?

Get help, child.

netperf-wrapper from bufferbloat.net

[mtaht]

Over at bufferbloat.net we have developed several pretty accurate bandwidth and latency measurement tests, that work at speeds up to 40GigE. We wrap the popular with the linux-netdev's "netperf" tool with something that can aggregate and plot the results, called "netperf-wrapper". The most popular test in the suite is called "rrul" which stands for "Realtime Response Under Load", but there are many others in the suite. It has been used to extensively tune several fair queuing and aqm algorithms, notably "fq_codel" which is in cerowrt, openwrt, and many other 3rd party firmwares. Its been used to debug network hardware, wifi, cablemodems, and most recently during the 40GigE batch-bql patchset now entering the linux kernel. Some examples of use to tune a smarter queue management system against modern day cable modems: http://burntchrome.blogspot.co... http://snapon.lab.bufferbloat.... There are also netperf-wrapper results for 40GigE, DSL, and wifi spread around the Internet. The intermediate format netperf-wrapper uses to store its results are in json and parsable by anything, and I keep hoping someone will get around to writing a web interface for the datafiles... Nothing else I've ever seen even comes close to netperf-wrapper for finding good, accurate, long term numbers and multiple forms of anomoly. Pretty much all the web based tests get increasingly inaccurate above 20Mbits. Single threaded TCP tests are bad also as they generally result in someone defeating TCP congestion avoidance in pursuit of the best benchmark numbers. [2] Far more important to the debloaters is not the bandwidth attained but the latency induced while getting it. [1] We maintain several public servers for netperf-wrapper, all connected via a gigE connection to the internet. Thus far we haven't overloaded them (nor advertised them widely), but if you want to give netperf-wrapper a try, and can't set up your own netperf server on the other side, feel free to ping us on the bloat mailing list for some addresses on various continents. [1] A brief rant: Bandwidth != speed. Bandwidth is capacity/interval. Real perceived speed is obtained via low latency. [2] I really hate that all the web network measurement tests don't simultaneously measure ping while running their upload and downloads. IF ONLY those tests would do that, people would start to realize that there is a huge tradeoff between good latency and high bandwidth, and that they are doing their networks in, by optimizing for bandwidth only. Networks engineered for speedtest's current test, *suck* for voip and gaming. I'd like to petition them to at least report ping times under load to the 98th percentile.

Re: Speakeasy Speed Test

[Dishevel]

I have to tell you that the stuff that normally kills a fiber run leaves copper alone and vice versa.

If you are running a company and a connection is important spending an extra grand or so a month to have bonded T1 backup is a good idea.

Pakistan Broadband Ptcl Speed Test

[SyedaAyesha]

if you are in Pakistan and want to check your exact broadband ptcl speed test check at http://www.pakistanihub.pk/201...

Answered your own question

[Vrtigo1]

The speeds I experience are much, much slower than the speed tests show I'm capable of.

You answered your own question...if you're looking for a more accurate test that shows the speed you're likely to experience through the course of your normal browsing, then why not just use your normal browsing as the test? There is really no other accurate option since the Internet by its very nature is decentralized so even if you found a non-prioritized speed test server, those results would be meaningless for any other hosts because your traffic would likely flow a completely unrelated path to reach them. The ISP speedtest sites are usually located within the ISP's own network so they're a good measure of throughput from you to the edge of the ISP's network, which is all they really care (or can care) about. Just use a different ISP's speed test site if you want results from outside your ISP's network.

Find a real data source you can trust,not a 'test'

[almondo]

I have 90Mbit up * 10Mbit down lightning service from Brighthouse and it is quite real. I can say for certain it is real because I have a co-located machine at Terramark on a 1GBit link running SNMP and I move enough data both ways to be able to do the math to validate. The fact is that they deliver well over rated speeds as well as I routinely push 11Mbits sustained up and pull over 100Mbits sustained down. Sustained to me means over at least an hour down with some of
my sustain up running 8 or more hours(a lot of cameras, a lot of data pushed offsite everyday).

One thing you really need to understand in this battle for bandwidth is that you are absolutely owned by your network transit path. An interior network (you are part of your ISP's interior network in the context I refer to) may have plenty of capacity while their edges may be grossly inadequate (as in Comcast and AT&T the last time I was on their pipes) and this fact can thoroughly convolute your test results because they can (and some definitely do) divert bandwidth test traffic onto a better path than you will ever see with real traffic.

The short answer IMHO is that you can only really determine true bandwidth with a real, uncongested validation point that you can trust. Bandwidth tests are circumvented other ways too. One trick is traffic shaping with a burst that gives you full rated pipe for a minute then hacks you down step by step until you get what they decide you get sustained. That will show high bandwidth in a test but the ISP chosen rate will surface when you actually move some traffic around.

Personally, (and Larry Ellison may want to kill me for this) I have used various Oracle image downloads (not little Java tarballs but ISOs for Solaris and other various big Oracle stuff) as a basis for occasional test in the past. My trust in this methodology stems from the fact that I can routinely pull over 300Mbits from Oracle to my co-located host and I can nearly always saturate my inbound to well above spec on Brighthouse.

Re:Find a real data source you can trust,not a 'te

[almondo]

Sorry, it is actually 90down * 10up.

Could be your wifi

[MrResistor]

I was having similar problems (Uverse though, not Comcast). On a lark, I dug an old Ethernet cable out of storage and ran it from my gateway router to my desk. Problem solved.