Thinkbroadband Speed Tests

I am supposedly moved from the BT FTTC 40/10 offering to the 80/20.

Does anyone know a possible reason why I consistenly get better throughput on this site's flash speed test thatn I do on the 'normal' tester.

Flash gives me 46 / 10 instead of 35 / 4.

Could there be something screwy on my Pc as to why the http stuff is slower ?

t&r

The normal as in the Java versus Flash?

In reply to a post by bathsheba666:

Flash gives me 46 / 10 instead of 35 / 4.

Could there be something screwy on my Pc as to why the http stuff is slower ?

It's a known issue. In the sense, at least, that several of us know about it. In my case neither test returns correct speeds for the upstream. My 80/20 connection is reported as 70 or higher down by all testers and 15 or higher up by all testers except TBB which claims 8Mb/s.

In reply to a post by Andrue:

It's a known issue. In the sense, at least, that several of us know about it. In my case neither test returns correct speeds for the upstream. My 80/20 connection is reported as 70 or higher down by all testers and 15 or higher up by all testers except TBB which claims 8Mb/s.

Likewise - TBB Flash tester consistently gives higher results than Java tester, particularly upload, which never exceeds about 8Mbps on TBB Java whereas pretty much all other testers show 17.

Here are mine just now:



The TBB flash test gave 70.9 Mbps download, 17.66 Mbps upload. (The download is a bit slower than speedtest.net but the flash test upload is a little higher.)

The daft bit is we know that Java can got a lot faster

http://www.youtube.com/watch?gl=GB&feature=channel_p...

1m 37 seconds, over 250 Meg for both download and upload, on a fairly slow Apple some four years ago.

In reply to a post by MrSaffron:

The daft bit is we know that Java can got a lot faster

I've seen faster on a corporate connection.

We have seen faster too - that test was four years ago on what was an oldish laptop at that time

What does this test make of your connection ? It's the only one that gives me consistent results I would have expected from my service.

In reply to a post by MrSaffron:

The daft bit is we know that Java can got a lot faster

http://www.youtube.com/watch?gl=GB&feature=channel_p...

1m 37 seconds, over 250 Meg for both download and upload, on a fairly slow Apple some four years ago.

It is a puzzle. It seems to be connection or ISP specific though. I have three computers here and they all report roughly the same speed. That's a laptop and a low power server running Win 7 and an old desktop running XP. And yet if it's the connection how come it hits TBB - and only on the upload?

Weird.

In reply to a post by Andrue:

It seems to be connection or ISP specific though.

Do we have any comparisons from TalkTalk or Sky FTTC users yet? It would be interesting to rule out WBC, and MSILs etc.

Edited by jchamier (Mon 17-Sep-12 19:43:29)

THanks.

It gives 49.1 over 10.2 with a 99% rating.

Correct

The issue isn't Java, but Thinkbroadband's java tester (or maybe the server backing it up).

The most reliable speedtest I have found is http://mcslhr.visualware.com/myspeed/myspeed_line_ca... . It's the only tester that seems to get upload speeds that reflect a good real world site.

Just noticed Zarjaz referenced the same test above. For me it reports 68.6Mb download and 17.0Mb upload (Infinity 2). I understand I could increase upload a little by turning off QoS on the modem. I stopped using Thinkbroadband test when I moved to Infinity; completely useless now. Just reported Speed Down 31483.66 Kbps ( 30.7 Mbps ) Speed Up 3999.75 Kbps ( 3.9 Mbps )

Edited by StephenTodd (Mon 17-Sep-12 19:55:40)

If it was the server why on corporate fast connections does the same server give decent speeds?

What is going on is

1. Some ISPs and traffic management, we don't use HTTP protocol
2. Possible crude per thread management in the WBC network, or BT Retail segment
3. Some testers showing burst speed (or peak'ish)

The updates to the Android tester give some hints for the future

The speeds from the visualware site report figures very close to sustained values I get with 'real' downloads and uploads from good quality sites. (I'm happy to say that mostly I have seen sustained speeds very close to burst speeds on Infinity).

Just tried Thinkbroadband on port 80, same sort of results. Speed Down 32974.94 Kbps ( 32.2 Mbps ) Speed Up 4027.43 Kbps ( 3.9 Mbps ). Maybe your tester uses something seen by BT as p2p?

In reply to a post by StephenTodd:

The issue isn't Java, but Thinkbroadband's java tester (or maybe the server backing it up).

I trust Seb and MrSaffron to know the network the tester is connected to. The video shows in 2008 it could do 200+ Mbps upload and download - and I've seen 80Mbps upload and download on a corporate connection with AT&T a few years ago from the java tester.

The most reliable speedtest I have found is http://mcslhr.visualware.com/myspeed/myspeed_line_ca... . It's the only tester that seems to get upload speeds that reflect a good real world site.

For home broadband - using the BTwholesale WBC network. I'm still looking for reports from Sky and TalkTalk fibre users, or Hyperoptic users, or Digital Region (assuming they're not using WBC) that either have the same issue, or don't have the issue.

Just noticed Zarjaz referenced the same test above. For me it reports 68.6Mb download and 17.0Mb upload (Infinity 2). I understand I could increase upload a little by turning off QoS on the modem. I stopped using Thinkbroadband test when I moved to Infinity; completely useless now. Just reported Speed Down 31483.66 Kbps ( 30.7 Mbps ) Speed Up 3999.75 Kbps ( 3.9 Mbps )

It does sound as if BT Broadband's Infinity product is doing something odd. My installation is Weds.

In reply to a post by StephenTodd:

Maybe your tester uses something seen by BT as p2p?

That I think is what MrSaffron has been alluding to - the problem is in the ISP's management. Not every ISP does traffic management as you're aware being a former BEing.

Very much so, as in our protocol is dealt with as unrecognised which means it gets lumped into the p2p cluster a lot of the time.

In reply to a post by MrSaffron:

Very much so, as in our protocol is dealt with as unrecognised which means it gets lumped into the p2p cluster a lot of the time.

Quite a nice illustration of how simplistic traffic management actually is, even worse when it tries to identify based on content.

In reply to a post by jchamier:

In reply to a post by StephenTodd:

Maybe your tester uses something seen by BT as p2p?

That I think is what MrSaffron has been alluding to - the problem is in the ISP's management. Not every ISP does traffic management as you're aware being a former BEing.

I'm with IDNet. I think it would come as a surprise to a lot of people if IDNet had any kind of throttling implemented. If they do then surely it would be on the downstream not the upstream since the upstream doesn't count against your allowance. It is still WBC of course but I just want to clarify that I at least aren't on BT Infinity. Mind you I also get poor results from the flash tester.

Edited by Andrue (Mon 17-Sep-12 20:59:51)

In reply to a post by Andrue:

I'm with IDNet. I think it would come as a surprise to a lot of people if IDNet had any kind of throttling implemented. It is still WBC of course but I just want to clarify that I at least aren't on BT Infinity. Mind you I also get poor results from the flash tester. I'll see if I can run the Android one.

Yes, that's why I'd like to rule out WBC. It could be "big scale" traffic management at the WBC level between PPP "virtual circuits" - essentially is 100Gbit/s of IDnet traffic being slowed down slightly in preference to another WBC customer. All hidden in the massive WBC "unknown".

In reply to a post by jchamier:

In reply to a post by Andrue:

I'm with IDNet. I think it would come as a surprise to a lot of people if IDNet had any kind of throttling implemented. It is still WBC of course but I just want to clarify that I at least aren't on BT Infinity. Mind you I also get poor results from the flash tester. I'll see if I can run the Android one.

Yes, that's why I'd like to rule out WBC. It could be "big scale" traffic management at the WBC level between PPP "virtual circuits" - essentially is 100Gbit/s of IDnet traffic being slowed down slightly in preference to another WBC customer. All hidden in the massive WBC "unknown".

I just installed and ran the Android app. The download was reported low (around 20Mb/s) but was all over the shop and I think that's down to my wifi. I switched to wired several months ago because of a lack of channels. But most interestingly the phone upload was fine. It settled down reasonably quickly...to 14.9Mb/s.

So it's not my connection. It's my Windows computers. Or their OS. Or..something.

Edited by Andrue (Mon 17-Sep-12 21:10:07)

In reply to a post by Andrue:

So it's not my connection. It's my Windows computers. Or their OS. Or..something.

Or oracle have broken something in java since buying Sun ?

My first guess would be to actually uninstall and reboot any security software products. Not something most people can do without a spare PC.

In reply to a post by jchamier:

In reply to a post by Andrue:

So it's not my connection. It's my Windows computers. Or their OS. Or..something.

Or oracle have broken something in java since buying Sun ?

My first guess would be to actually uninstall and reboot any security software products. Not something most people can do without a spare PC.

Yeah. Although I'll note that the desktop upstairs is owned by my employer and has McAfee AV whereas the laptop and server are running Avast. It would be hard to find many points of similarity between those two machines.

In reply to a post by jchamier:

In reply to a post by Andrue:

So it's not my connection. It's my Windows computers. Or their OS. Or..something.

Or oracle have broken something in java since buying Sun ?

My first guess would be to actually uninstall and reboot any security software products. Not something most people can do without a spare PC.

Okay the plot thickens - sorta

My work machine no longer wants to run the Java test. Doesn't dispaly the loading circle nor a message about installing Java. Doesn't do anything. The Flash tester runs but produces a poor download figure (30Mb/s as compared to my laptop and server that return 70Mb/s or better) but reports 14.9Mb/s upload. So that does suggest some kind of software interaction on my laptop. FWIW on the laptop the Flash based tester gives really bad results of around 5Mb/s upload.

Or oracle have broken something in java since buying Sun ?

Java v7 came out with some defaults going to ipv6 rather than ipv4. This caused me and several other people quite a problem running Java based tests and other Java applets: the BT tests routinely failed to run and stuck at 96%, for example. This isn't the reason for the lower speeds on the Thinkbroadband tests however; I think it must be as others have suggested that BT (or someone in the path) is interpreting the traffic as p2p and throttling it.

In case it helps anyone with other Java issues, try going to the Java control panel, navigate to the Java tab, click View, and under your Java v7 insert a runtime parameter of -Djava.net.preferIPv4Stack=true; or disable ipv6 in your network settings.

The test running slowly is of course not important of itself, but might be a useful indicator/diagnostic for other applications that do not behave properly under Infinity etc.

Edited by StephenTodd (Tue 18-Sep-12 09:35:28)

In reply to a post by jchamier:

In reply to a post by Andrue:

It seems to be connection or ISP specific though.

Do we have any comparisons from TalkTalk or Sky FTTC users yet? It would be interesting to rule out WBC, and MSILs etc.

I have Sky fibre and here's mine from the TBB Java speed test:


The Flash one gives me 45.8/17.6 and the speedtest.net one gives me 57.1/15.7. The line is sync'd at 59999/20000.

In reply to a post by simon194:

The Flash one gives me 45.8/17.6 and the speedtest.net one gives me 57.1/15.7. The line is sync'd at 59999/20000.

Thanks! That seems to answer it, the tbb Java tester shows slow uploads for FTTC even on ISPs that claim no throttling or traffic management.

It explains some things but not all.

I see (Infinity) down/up
TBB Java 30.2/3.8
TBB Flash 37.34/15.56
iPlayer diagnostics 63 (download only)
speedtest.net 68.17/15.47
capspeed 67.8/16.8

The upload loss on TBB Java is much more than others see, and even TBB Flash is well below other testers.

In reply to a post by StephenTodd:

The upload loss on TBB Java is much more than others see, and even TBB Flash is well below other testers.

Interesting. Do you use tbbMeter?

On BE ADSL2+ I get the following, the colour descriptions are what I see in the tbbMeter graph.

ADSL sync = 16,759 kbps / 1,209 kbps

tbb flash Download = 13.81 Mbps Upload = 0.97 Mbps
tall solid green block / tall but thin red line

tbb java Download = 13.6 Mbps Upload = 1 Mbps
jagged green block / short red/yellow block

speedtest.net / newbury
Download = 14.33 Mbps Upload = 1.02 Mbps
tall solid green block / short red block

Pic of TbbMeter:
http://i612.photobucket.com/albums/tt201/MisterReder...