[Update] ALL 20>30Mb upgrade discussion

[Peter_]

Originally Posted by qasdfdsaq

It can be more accurate, but it can also be less so. And while I'll agree the instructions are fine, and no speed test is definitive, that particular test is much more vulnerable to being skewed by factors such as the user's machine and the timing of the user's actions. Having to manually add up four numbers that *will* be actively changing and counting on an end-user to accurately do math adds extra variance factors. It might give you a ballpark figure, but that's it, and it's definitely prone to over-estimating speed rather than underestimating, which may be why VM like it so much...

It gives similar results to every other speedtest that I have used with my machine even ones not in the public domain.

But does it matter if one is getting 26.8Mb/s rather than 30Mb/s?

[Peter_]

Originally Posted by Hugh

But does it matter if one is getting 26.8Mb/s rather than 30Mb/s?

As it is classed as an up to figure and you are getting 27Mb it may not be classed as an issue as it is well within range for 30Mb, but I know you have had that speed since install so maybe a call to support on Monday would be in order as the may well be a fault that needs sorting out.

The config is set at 33Mb and many like myself do get over 30Mb.

I am thinking of your replacement modem while you were still on 20Mb and you still have the same issue, I would definitely call up and get them to check the connection.

[qasdfdsaq]

I think we need to get past the constant "it works for me" attitude. Just because it works for you doesn't mean it works for everyone, or that it isn't fundamentally flawed.

It has too many variable factors - for one, how your browser average download speeds, how fast a user clicks "Save", etc. to be considered on any level a scientific test. Other factors will skew it even more.

I just did the test in Firefox, 43mbps with one file, 52mbps with 4 in Firefox. With Internet Explorer, it gave me 64mbps, that's before adding on the 10% they tell you to - do that and it's 70mbps, a result that's clearly impossible on my 50mb connection, while following the same instructions, and while actually transferring data at the same speed. Why? IE8 shows average speeds for the whole download, Firefox (4b11) only shows the average of the last few seconds. This alone makes results incomparable, that's without incorporating factors such as how fast a user clicks or computer performance. IE6/7 behaves differently again, with precaching and stuff going on that wildly skews numbers even further. I could probably get it to say 100mbps if I clicked as slowly as some of my colleagues do in the office...

Like I said, ballpark figure to figure out if the connection's working? Sure. Any sort of accurate result even close to commercial speed tests? Not a chance in hell.

---------- Post added at 23:46 ---------- Previous post was at 23:46 ----------

Originally Posted by Hugh

But does it matter if one is getting 26.8Mb/s rather than 30Mb/s?

Does it matter if the test says 70mbps on a 50mb connection? That's how far off it is.

Sure, 26.8 out of 30 is fine. I wouldn't complain about that. But a test that will overestimate by that much I cannot and will never trust.

[Edit]
Just did it again, IE8, following instructions to the letter, but left downloads for 45 seconds instead of 30. Cleared browser cache, started 4 downloads of >400MB files, left downloads "to settle" for 45 seconds and took a screenshot to snapshot speeds. 57.84mbps "data related to the files being transferred". Add on 10% as they instruct, that'd be 63.62mbps.

Now I'm a technically competent person with a fast computer, no firewall or AV (temporarily disabled) and a connection that's pretty much working perfectly bang on it's correct speed right now. If the test gets things this wrong for me, how useful could it ever be for your average layman?

[Chrysalis]

I am also curious why they need to download 4 at once, I can get 3.8meg/sec on one file only. The only reasons I would think of needing to do 4 are.

1 - to hide congestion as multi threaded in congestion conditions gives a bigger share of bandwidth.
2 - to hide rtt variance (jitter), if base latency is also high then it would mean need a higher tcp rwin to sustain higher speeds single threaded.
3 - if on windows XP or older with a small default rwin setting. (on gigabit lan it defaults much higher tho).

[qasdfdsaq]

Probably both 1) and 2).

Samknows data already shows VM is more affected by single vs. multithreaded tests (gains most from going from 1 to multiple download streams) and have the highest jitter.

Plus as I say, the using multiple files in itself inflates numbers with Internet Explorer.

That said, a lot of advertising about faster broadband speeds these days has tended towards "do more things at a time" rather than "do one thing faster" as companies find it harder to maintain one fast connection vs. several slightly slower ones. If network performance isn't sufficient to fill the pipe with one connection (which is becoming the case more and more often, not always the fault of the ISP mind you) then several will fill the pipe better, but only to the detriment of other users.

Essentially if there isn't enough bandwidth for everybody, running more downloads means you end up taking proportionally more, partly masking the problem. And gives an artificially inflated speed with what is still the most common browser. No wonder VM like it...

[darkm]

Can confirm that speedtest.net is giving false readings at the minute, download speed is fine

[Peter_]

Originally Posted by Chrysalis

I am also curious why they need to download 4 at once, I can get 3.8meg/sec on one file only. The only reasons I would think of needing to do 4 are.

1 - to hide congestion as multi threaded in congestion conditions gives a bigger share of bandwidth.
2 - to hide rtt variance (jitter), if base latency is also high then it would mean need a higher tcp rwin to sustain higher speeds single threaded.
3 - if on windows XP or older with a small default rwin setting. (on gigabit lan it defaults much higher tho).

It balances out the download because if one file is downloading very slowly but the other is flying that indicates an issue, whenever I download all files are fairly well balanced.

[Chrysalis]

that test will always report file 1 fastest file 2 next 3 3rd and 4th slowest because when file 1 starts it will have max speed, when file 2 starts its sharing with file 1 and so on, the browser reports average not live speed.

unless of course can start all 4 at exactly the same time somehow like using a download manager.

[Peter_]

Originally Posted by Chrysalis

that test will always report file 1 fastest file 2 next 3 3rd and 4th slowest because when file 1 starts it will have max speed, when file 2 starts its sharing with file 1 and so on, the browser reports average not live speed.

unless of course can start all 4 at exactly the same time somehow like using a download manager.

I have no idea why you would think that as they should not all balance out because if you try the link they will all start to match up to the same speed.

You will get the same result using Firefox or Internet Explorer.

[qasdfdsaq]

Originally Posted by Masque

I have no idea why you would think that as they should not all balance out because if you try the link they will all start to match up to the same speed.

You will get the same result using Firefox or Internet Explorer.

Incorrect on both counts, as I've already demonstrated.

[Chrysalis]

Originally Posted by Masque

I have no idea why you would think that as they should not all balance out because if you try the link they will all start to match up to the same speed.

You will get the same result using Firefox or Internet Explorer.

rethink what you said, and you work in tech support?

obviously the results will be skewed as before the 2nd link is clicked the first download has full line capabilities to itself.

eg. I just did it just now. I left a 10 second gap between each click.

file 1 2.1meg/sec
file 2 1.6meg/sec
file 3 1.2meg/sec
file 4 1.1meg/sec

so my line can do 6.0meg/sec on a 30mbit connection. impressive incidently file 4 started about 0.9meg/sec but when earlier files finished it then had full line speed bringing up its average at the end.

[Peter_]

Originally Posted by Chrysalis

rethink what you said, and you work in tech support?

You need to rethink what you are saying as your interpretation is completely wrong of have this test works.

Originally Posted by Chrysalis

obviously the results will be skewed as before the 2nd link is clicked the first download has full line capabilities to itself.

eg. I just did it just now. I left a 10 second gap between each click.

file 1 2.1meg/sec
file 2 1.6meg/sec
file 3 1.2meg/sec
file 4 1.1meg/sec

so my line can do 6.0meg/sec on a 30mbit connection. impressive incidently file 4 started about 0.9meg/sec but when earlier files finished it then had full line speed bringing up its average at the end.

I do not think you actually understand what is meant to happen which is pretty obvious with your comments.

On my machine they each balanced out downloaded together at around 640kbps per file and the were 6 files, time is irrelevant as we want them to balance out over the 6 files and then you add them together and add 10% for anything in the background.

[qasdfdsaq]

Now you've had two people clearly prove you wrong with evidence, I think you're the one who needs to rethink how the test works.

They will not balance out as they will be run over different times, which are not completely overlapping. Time is completely relevant.

I like how you've completely ignored how I've already explained this and shown you it in practice.

[Ignitionnet]

Originally Posted by Chrysalis

rethink what you said, and you work in tech support?

obviously the results will be skewed as before the 2nd link is clicked the first download has full line capabilities to itself.

eg. I just did it just now. I left a 10 second gap between each click.

file 1 2.1meg/sec
file 2 1.6meg/sec
file 3 1.2meg/sec
file 4 1.1meg/sec

so my line can do 6.0meg/sec on a 30mbit connection. impressive incidently file 4 started about 0.9meg/sec but when earlier files finished it then had full line speed bringing up its average at the end.

You're supposed to have a snapshot of the speed of all of them, once settled down, when all 4 are downloading simultaneously not their average speeds when all done. That would indeed be a futile measure.

Again all 4 must be running at the time you check their speeds, and all 4 must have had time to ramp up.

---------- Post added at 12:35 ---------- Previous post was at 12:34 ----------

Originally Posted by qasdfdsaq

Now you've had two people clearly prove you wrong with evidence, I think you're the one who needs to rethink how the test works.

They will not balance out as they will be run over different times, which are not completely overlapping. Time is completely relevant.

I like how you've completely ignored how I've already explained this and shown you it in practice.

See above - it's supposed to be a snapshot of when all files are downloading simultaneously and have had the chance to ramp up.

Obviously not ideal but you are getting the wrong impression if you are looking at speed at the end of the download.