Peak versus off-peak speeds

Interesting, a look at speed test data for the big six providers

http://blog.thinkbroadband.com/2015/07/what-differen...

Probably not what some of you were expecting for PlusNet either.

any idea why almost all providers have a speed drop in the last 3 months or so

Someone is having a larf! Or taking a wee-wee.

a question - since I can't remember - are those single thread tests or multiple thread tests as Plusnet problems are on single thread tets

Multi thread - thought we had seen enough people moaning about the multiple thread.

Remember single thread might just be an effect of the traffic management i.e. unknown protocol on an odd port.

From the posts on the Plusnet Forums - it appears that they were all complaining about the rotten single thread results

Could blame

1. G.INP confusion
2. Increased cross talk
3. Not all providers have the same degree of dip and one or two have gone up - just not those
4. Did look at ratio of PC to phone to tablet and pretty stable for a few months
5. The Which? campaign which did boost visitors may have got people on slower lines to test
6. People downgrading from FTTC to ADSL2+ to save a pound or two
7. People downgrading from up to 76 to up to 38 Mbps
8. Its not us, did check

My feeling from looking at all the data is its a slight shift in our demographic, i.e. more ADSL2+ users versus fibre combined with some dips in speed, i.e. moans are not confined to just one provider.

Ok it was on the todo list already, so here is a beta for a version of the new test that uses just a single thread

http://labs.thinkbroadband.com/speedtest/?site=omega...

Looks the same as the new web version, but slightly different text on start page, and if you share the graph at the end you can spot difference as it shows a tbbx1 result. The page that does the graph drawing will need some tweaking in the future, but as hinted elsewhere waiting on some other stuff.

These tests are logged under a specific version, hence the random string, so if the consensus is that this does not work then we can exclude from the monthly results. If it does work then after a decent period of data collection will look at how adding a single thread run before or after a multiple thread run.

My own connection non-PlusNet has been showing single thread problems today, so a good day test/code this version. Blames Wimbledon and people streaming while at their office desk.

In reply to a post by Oldjim:

it appears that they were all complaining about the rotten single thread results

Mine's been pretty rotten p.m. recently never getting past the single threaded part. Can't get much worse than a big fat zero - can't go backwards can it 8^)

clarify those are 6x thread stats or single threaded?

I trust the source that revealed circa 10% of plusnet's customer base is affected by current BTw issues.

All i can say here is for me personally plusnet's performance has significantly regressed during 2015, stats printed to try and point out I am alone in this experience doesnt help the matter.

Plusnet even confirmed themselves they had congested endpoints a couple of months back, as I understand that issue got fixed tho.

then those results are useless, please print the single threaded. This is important andrew, its been made clear the issues affect single threaded performance. I hope you are as keen to show the single threaded tests as you were the multi.

Edited by Chrysalis (Tue 07-Jul-15 22:11:21)

That was httpx6 in the blog

Need more data so more registrered users doing speed tests basically

The beta web single one might get used once I've had time to look and plusnet users confirm it behaves the same way as the flash, need confirmation due to the way traffic management operates

Do you have any individuals' data that tally or contradict the graphs?

I ran this one but it only did one download test which is single or multi threaded?

http://labs.thinkbroadband.com/speedtest/?site=omega
http://www.thinkbroadband.com/speedtest/results.html...

The main difference will be on the upload test, on html5 it uses the browser specified sendbuffer, which e.g. firefox is the network.tcp.sendbuffer variable.

Flash will use the OS send buffer.

Edited by Chrysalis (Wed 08-Jul-15 00:50:05)

Since I joined Plusnet I have had very consistent speedtests, with almost completely flat lines on the graph, e.g.
http://www.thinkbroadband.com/speedtest/results.html...

When I was on Infinity the graph jumped up and down like mad, especially the HTTPx1 test.
Consistency in the Plusnet test is probably helped as I am on 40/20, synced at 50 with available speeds well above 70.
I had 80/20 on Infinity, so the bouncing HTTPx1 test still usually averaged above 40.

I just tried the other test at
http://labs.thinkbroadband.com/speedtest/?site=omega
and the upload on that zigzags like mad, though keeping pretty much to the same overall speed
http://www.thinkbroadband.com/speedtest/results.html...

Probably not really significant, and maybe just due to different sampling/smoothing on the result data

Yes different methodology for the upload so will never be as flat as the flash on that side

Also that was the multi thread you ran as the site parameter was curtailed.

Edited by MrSaffron (Wed 08-Jul-15 09:24:34)

As the graph shows two sets of download figures it did the HTTPx1 test, as I said the speed test result image needs tweaking to reflect this beta.

The vast majority probably don't want another 12 seconds added to the duration of the speed test, so single threading may end up being something you have to choose as part of diagnostic testing etc

As you know in large datasets there are always individuals that contradict the overall results. If everyone was a model connection we could just speedtest one user from each ISP and publish a big table 100+ ISP long.

In reply to a post by MrSaffron:

Ok it was on the todo list already, so here is a beta for a version of the new test that uses just a single thread

http://labs.thinkbroadband.com/speedtest/?site=omega...

Looks the same as the new web version, but slightly different text on start page, and if you share the graph at the end you can spot difference as it shows a tbbx1 result. The page that does the graph drawing will need some tweaking in the future, but as hinted elsewhere waiting on some other stuff.

These tests are logged under a specific version, hence the random string, so if the consensus is that this does not work then we can exclude from the monthly results. If it does work then after a decent period of data collection will look at how adding a single thread run before or after a multiple thread run.

My own connection non-PlusNet has been showing single thread problems today, so a good day test/code this version. Blames Wimbledon and people streaming while at their office desk.

11am on Zen

http://www.thinkbroadband.com/speedtest/results.html...

I have noticed it seems to be servers located in London. I get the same results on speedtest.net from London servers but go elsewhere and it's fine.

http://www.speedtest.net/my-result/4489332711

Edited by deleted (Wed 08-Jul-15 11:17:57)

So is the test linked above a single thread test or multi thread test?

http://labs.thinkbroadband.com/speedtest/?site=omega... is a single thread test and it says so in the welcome message

How do you know where the server is on our test?

speedtest.net is always multi thread too by the way

Sorry my confusion is showing here. I thought I saw a video of Seb in London and he was using your server to demonstrate it's speed?

no idea where your server is - thats your business

The files used are currently in London, but nothing stopping me switching them to almost anywhere and for some testing in the past we have used multiple locations.

So do not assume file location.

In reply to a post by MrSaffron:

As you know in large datasets there are always individuals that contradict the overall results.

Of course. I was hinting that you could use your own and other 'trusted' connections to attempt to ascertain if the median/mean results are typical across these trusted connections. Try to cut out the local connection issues (or point to them, if applicable).

With regards Plusnet, my guess is that the traffic management angle is part of it. Perhaps Plusnet doesn't suffer from a higher percentage of heavy users in the same way others do, and the habits of said users may create large amounts of night-time traffic.

I'm on a legacy product and from time to time, I have to consider after midnight for High Def or excessive downloads. Maybe the heavy users were 'trained' to behave in a similar fashion prior to unlimited conmtracts, and with PN's FUP they avoid PN.

If all BB providers publicly gave their capacity data, we could probably guess more readily.

Andrew you seem hesitant with single threaded tests.

1 - Can you publish the single threaded results, you have some, even if its a small set of users it is useful to see.
2 - I dont see the issue of doing single threaded testing on guest speedtests, if you dont want to do both then do single threaded only, much more useful data.

If you revert to multi threaded only, then you just another clone of speedtest.net etc. TBB is unique in the single threaded speedtesting, but I am very disappointed that information doesnt get published, and seems was a mistake to not have it on the guest speedtests.

wow what an arrogant response

I'll assume what I like thank you very much. Is it my fault on the xmas tree lights YouTube video you look like a cockney muppet who never has been or ever will be laid ?

Go [censored] yourself! (don't need you or your site anyway!)

We have always said single threading will becoming back and this test is part of that journey.

Enough people complained about wrong results when guest version showed both sets, and also about the time the test takes, with enough abandoning early on that we would not have a decent data set from a user.

I only have so much time to do changes and analyse results, so you tend to get stuff published as soon as I am happy.

Have left the previous reply in place, so people can understand why I am saying no warnings you are suspended.

Attacks on staff are not tolerated.

"Attacks on staff are not tolerated. " - absolutely agree. Unforgivable, but that would be the case also if it were another user wouldn't it?

How is it possible for a 'burst' which is higher than the line speed?

Because of the way various buffers operate in the hardware and software of the computer, including AV and firewall software.

So basically, all speed tests are an approximation?

More that all speed tests can be affected by other factors. Speed tests can't necessarily control things like caching. Theoretically a network device could do all the requests for data and dump them down to the client in one go in which case the speed test would have nothing for a while and then suddenly a massive spike.

EDIT: The speed test is actually testing every component in the chain and therefore it shows where individual components could be doing something to the traffic. You would have to remove all of those components to get a "pure" test.

Edited by ian72 (Wed 08-Jul-15 16:37:40)

Thanks for the clear explanation.

In reply to a post by wingco1:

So basically, all speed tests are an approximation?

As the others have said, depends on buffering etc. A few years ago it was very common for Kasperksy users to get download estimates many times faster than their lines: eg 120Mbps on a 20Mbps line. This was down to some curious interference that Kaspersky did (possibly to get big enough buffers to do a reliable check) that interfered with the 'natural' flow of the data and broke the assumptions made by speed testers at the time. The interference did not a make significant difference to real data on the line, somehow just to speed tests.

Edited by StephenTodd (Wed 08-Jul-15 18:09:28)