|
|
Most speedtests just do a single ping test before checking the download and upload speeds, but the new speedtest from DSLreports constantly checks latency during the test, providing a nice graph of a connection's responsiveness under load.
Here's two tests i took earlier:
With fq_codel and SQM scripts active on my router
Without fq_codel and SQM scripts active
Hopefully more speedtests (including TBB's) will move towards including this important measurement in their tests.
Edited by deleted (Thu 30-Apr-15 16:16:45)
|
|
|
We've been collecting the data on buffer bloat for a number of months through our tester,
Plan was to simply the difference between idle and when downloading and when uploading, rather than another graph.
Being meaning to blog about it, but the coverage data updates and lots of other stuff has taken precedence. Have data from an ISP wide perspective already and makes for interesting viewing. In case someone thinks I'm bluffing here is the March 2015 data just posted onto twitter https://twitter.com/thinkbroadband/status/5938192551...
The difference between providers with respect to upload and download performance is interesting.
|
|
The author of the above post is a thinkbroadband staff member. It may not constitute an official statement on behalf of thinkbroadband.
|
|
|
http://www.thinkbroadband.com/speedtest.html
Will report a grade of buffer bloat for down and up directions.
|
|
The author of the above post is a thinkbroadband staff member. It may not constitute an official statement on behalf of thinkbroadband.
|
|
Register (or login) on our website and you will not see this ad.
|
|
|
Seems my router modem combination, DSL-N55U and Billion 7800 bridge is well bloated, I get a F on buffer bloat.
|
|
|
Which means if your latency was 40ms to get an F it had risen to six times that.
Grades were based on what looked like a nice answer from the ISP data. Time to see how it beds in, and get some FAQ entries up, Should have April data tomorrow to talk about too.
|
|
The author of the above post is a thinkbroadband staff member. It may not constitute an official statement on behalf of thinkbroadband.
|
|
|
|
|
|
|
I get an A up and down 
|
|
|
http://www.thinkbroadband.com/speedtest.html
Will report a grade of buffer bloat for down and up directions. I get B down and A up, which is about what I expect based on performance. Unfortunately, this tester seems to default to IPv6 if available, which in my case is a tunnel. Is there some way to force an IPv4 test other than disabling IPv6 on my computer or router?
Is there any intention to include the 'bloat grades' in the shareable results?
|
|
|
http://www.thinkbroadband.com/speedtest.html
Will report a grade of buffer bloat for down and up directions. Am I missing something or does this only work on Windows?
Can't see any reference to buffer bloat grades in either Linux (Firefox) or OS X (Safari & Chromium)
|
|
|
Can't see any reference to buffer bloat grades in either Linux (Firefox) or OS X (Safari & Chromium)
Log out of the main website before loading the speed test. That will load the HTML5 test.
Oliver.
|
|
|
Log out of the main website before loading the speed test. That will load the HTML5 test. Ah, that's better- thanks.
"A" in both directions
|
|
|
Clunkier than AOL 3.0 when doing downstream, Upload well it failed to even bother.
Grade F is a far cry from what DSLreports gave me
http://prntscr.com/7025ns
|
|
|
Well have changed nothing to do with the actual test, the A/B logic has been present for over two months. So if test has run for you before and not this time, something else was up
|
|
The author of the above post is a thinkbroadband staff member. It may not constitute an official statement on behalf of thinkbroadband.
|
|
|
Is the latency shown that achieved with or without the tests running? I guess it is WITHOUT as I get 10ms and F/D or F/E ratings!
Need to trying on another PC to see if this one, running XP and FF is the problem.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
M H C
taurus excreta cerebrum vincit
|
|
|
Latency is the figure from when no testing is underway.
Entirely possible to have an otherwise great connection and get poor ratings for the buffer problems due to CPE or a large buffer somewhere in the system.
|
|
The author of the above post is a thinkbroadband staff member. It may not constitute an official statement on behalf of thinkbroadband.
|
|
|
On TBB test I get B down and E up.
|
|
|
|
Excellent news! I'm glad to have the latency detection in the TBB tester.
When I turn off the SQM/fq_codel in my router, I get an F for bufferbloat. But when I turn it on, I get an A in both directions.
|
|
|
,,, In case someone thinks I'm bluffing here is the March 2015 data just posted onto twitter https://twitter.com/thinkbroadband/status/5938192551...
The difference between providers with respect to upload and download performance is interesting. Allowing that I don't know for sure what the figures on the charts mean (multiplying factors?), my ISP (IDNet) would appear, at best, to be no better than average- 4.9 up and 3.3 down.
My own tests (this morning, within a few minutes of each other) give an A both up and down for the tbb test and B on the DSLReports test, which I would assume is good.
So I'm not sure what the figures on those charts imply... That most IDNet users have rubbish routers? That a lot depends on the route taken to the tbb servers? Something else entirely?
Could you give a brief outline of how (or even if) these charts should be used in comparing ISPs or choosing one?
(Fair enough if you'd prefer to leave it until the FAQ is sorted.)
Edited by billford (Fri 01-May-15 17:44:51)
|
|
|
|
On my VM 100Mbps I get an F
On my BT 40/10 I get a C
I use a Draytek 2920 with the VM Superhub in Modem Only mode and the BT VDSL modem box plugged into the WAN ports.
I guess from this then that buffer bloat is a modem issue?
|
|
|
Need to do some digging ... other PC gives same results.
HG612 >> 2701router >> Cisco/Linksys managed switch >> PC(s) nothing obvious yet!
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
M H C
taurus excreta cerebrum vincit
|
|
|
|
It is a modem issue, but your Draytek also isn't doing anything to help.
Because the modems aren't likely to be fixed anytime soon (well maybe the Virgin one when they move to DOCSIS 3.1) you need to move the bottleneck into your router and solve the problem there. You do this by limiting the speed of your router's WAN connections to below their actual speed.
Unfortunately i don't know what facilities the Draytek 2920 has to help with this (after the trauma of running a 2820vn for a couple of years i've tried to forget the existence of Draytek all together).
|
|
|
|
Thanks - achievable I reckon. Will take a look later as I can put bandwidth restrictions in place.
|
|
|
|
|
|
The author of the above post is a thinkbroadband staff member. It may not constitute an official statement on behalf of thinkbroadband.
|
|
|
Its a bit of a complex one, but yes is largely down to the router people are using, hence why even fast providers can suffer it, while delivering otherwise good speeds.
With the Bring Your Own Router providers it is more complex but plan to try and show some profile charts for those providers in the blog where I have more space.
Grade A people should basically find that they can game while doing a lot with their connection, and there is a growing movement to get different algorithms embedded in CPE to help, though in the short term people can effectively do their own QoS on some devices.
|
|
The author of the above post is a thinkbroadband staff member. It may not constitute an official statement on behalf of thinkbroadband.
|
|
|
Its a bit of a complex one, but yes is largely down to the router people are using ... Hmm, interesting... pity you can't identify the router from your end 
|
|
|
Here's mine http://www.dslreports.com/speedtest/401304 do we live next door perchance ?
|
|
|
With the Bring Your Own Router providers it is more complex but plan to try and show some profile charts for those providers in the blog where I have more space.
Showing buffer bloat vs ISP is perhaps clouding the issue, when it's a CPE issue and not an ISP network issue. Some ISPs are going to have a much higher percentage of non-ISP supplied routers than others.
Short of creating a test page with a huge list of user-selectable routers in a menu though, it's going to be very difficult to graph buffer bloat vs router model. Perhaps a router brand drop-down is workable?
Oliver.
Edited by Oliver341 (Fri 01-May-15 12:50:00)
|
|
|
Maybe a drop down, but given when we've tried asking similar questions in the past the error rate is pretty high e.g. ask people if on 3G and you still get loads picking that when clearly on a fixed line provider.
All this extra need to explain is why it will be a blog item, so can include the profiles for each ISP, and then see patterns emerge and can then investigate further.
|
|
The author of the above post is a thinkbroadband staff member. It may not constitute an official statement on behalf of thinkbroadband.
|
|
|
|
Did some bandwidth restrictions and here are results (don't know what a - means better than an A or worse than an F?)
BT 40/10
100% C
75% B
50% B
25% B
VM 100/6
100% F
75% F
50% F
25% -
The VM is awful. So that either means it's modem or VM side.
Slight improvement on BT when reducing bandwidth useage.
|
|
|
|
On an FTTP connection from that well known commercial altnet I can report an 'A' both ways.
Typical ping times from the windows command prompt to google.com are 5ms.
|
|
|
What about the web browser?
If I run your speed test using Firefox the green line is above the yellow line, and I am tempted to think that I have an excuse to upgrade my Intel E6600 based PC.
If I use Project Spartan (from windows 10) the yellow line is above the green line.
I should say that I have moved from having one of the slowest connections (bottom 3%) to one of the fastest (top 3%)
Michael Chare
|
|
|
I think that's probably the best you're going to get using the Draytek. If you want better you'll probably have to switch to something that supports fq_codel (or nag Draytek to include support).
I typically limit my speed to around 90% of actual on both upload and download using OpenWrt and score an A+ on the DSLreports test and an A on TBB (both upload and download). During normal usage my latency doesn't increase by more than around 3-5ms and even stressing the connection using the RRUL test didn't see more than 10ms of additional latency on average (see my recent thread where i tested the Huawei and ECI modems).
Strange that you saw no improvement at all on VM though. Perhaps your area is just so congested nothing you do could help.
|
|
|
Link to the results please, verbal descriptions lack the necessary detail
|
|
The author of the above post is a thinkbroadband staff member. It may not constitute an official statement on behalf of thinkbroadband.
|
|
|
I got A down and F up. TalkTalk HG533.
That said, it's unusual for me to be uploading at full speed. Also, the HG533 does have comprehensive QoS settings which worked ok when I tested them.
Oliver.
|
|
|
Link to the results please, verbal descriptions lack the necessary detail
Firefox
Project Spartan
Michael Chare
|
|
|
At those speeds pc speed, browser and flash version have a large impact along with rwin.
Flash based test is unlikely to have lots of development time spent on it too
|
|
The author of the above post is a thinkbroadband staff member. It may not constitute an official statement on behalf of thinkbroadband.
|
|
|
can you post it somewhere else aside from twitter (image too small to read) and can you make this part of the speedtest results shown to the end user?
|
|
|
C here.I dont see it reporting for upload and download separate, just C overall.
I can see on the supplied graph tho I have more bufferbloat on uploading.
|
|
|
The dslreports test just gives a single combined mark, the tbb one gives separate up and down marks.
If you've got IPv6 you can run the tbb one without logging out of the main site by going to the Tools menu and running the IPv6 Speed Test.
later edit- but it's not the same test and gives poorer marks for buffer bloat 
Edited by billford (Mon 04-May-15 20:16:49)
|
|
|
now I am really confused can you please provide a direct url to the tbb test?
When I run the tbb speedtest I see nothing related to bufferbloat.
ok so have to logout sorry I didnt see that info.
I get a B for down and E for upload.
But the html5 test seems less reliable than the flash test, it 3 times during the test spiked up to 30mbit/sec although I know from dumeter actual throughput was a solid 19mbit/sec. This might be related to firefox stutters, as flash on firefox is protected from stutters due to been a separate process whilst html5 isnt.
Edited by Chrysalis (Mon 04-May-15 19:40:01)
|
|
|
ok I got it working now, are tbb planning to remove the single threaded tests? I hope not as they very useful.
Edited by Chrysalis (Mon 04-May-15 19:39:16)
|
|
|
now I am really confused can you please provide a direct url to the tbb test? It's the one on the orange button on the main site- but you have to be logged out (of the main site) or you'll get the standard Flash-based test.
The new one uses HTML5.
edit- it's the same link either way, it just takes you to one of two different places.
Edited by billford (Mon 04-May-15 19:45:53)
|
|
|
|
So can an ISP fix bufferbloat at their end of is it purely down to the users router if the problem occurs or not?
|
|
|
Important you set the lower advised P8 MTU in your router settings if you are suffering.
Edited by professor973 (Sun 07-Jun-15 22:27:50)
|
|
|
To be honest i rarely max out my connection so dont suffer that much. MTU is set to 1458 i think or whatever the Pulse8 activation email said
Edited by bobble_bob (Mon 08-Jun-15 07:01:04)
|
|
|
To the best of my knowledge, if a user's router suffers from bufferbloat then the only thing their ISP can do about it is to replace it with one that doesn't.
But I could be wrong.
|
|
|
Per user traffic management might help
|
|
The author of the above post is a thinkbroadband staff member. It may not constitute an official statement on behalf of thinkbroadband.
|
|
|
Probably cheaper to send them a decent router
|
|
|
To be honest i rarely max out my connection so dont suffer that much. MTU is set to 1458 i think or whatever the Pulse8 activation email said
I wonder if that's a hang-over from the issue with BT Wholesale several years ago. I'm not aware of any ADSL network that cannot run at 1492 MTU these days (including TalkTalk, who officially recommend 1432).
Oliver.
|
|
|
To be honest i rarely max out my connection so dont suffer that much. MTU is set to 1458 i think or whatever the Pulse8 activation email said
I wonder if that's a hang-over from the issue with BT Wholesale several years ago. I'm not aware of any ADSL network that cannot run at 1492 MTU these days (including TalkTalk, who officially recommend 1432).
Many use the PPP extensions to run at 1500 now as well, I can on PlusNet and did on BT both on FTTC.
plusnet unlimited fibre 80/20 product - Installed 2 June 14 - April Sync 57 / 11 with G.INP
16 years UK broadband (Since 1999 ntl:cable trial), Asus RT-AC68U & HG612 - BQM - Speedtest
|
|
|
Many use the PPP extensions to run at 1500 now as well, I can on PlusNet and did on BT both on FTTC.
Yeah, most networks are happy to run at 1500 MTU on PPPoA, with the notable exception of TalkTalk which is limited to 1492. Not sure how many networks run extensions to raise PPPoE to 1500, but that always struck me as overkill for the extra 8 bytes.
Oliver.
|
|
|
Well yesterday I got a new ac68.
The ac66 cannot handle pppoe at full FTTC sync speed without CTF, and it defenitly cannot handle it with QoS enabled.
My new AC68 can and now i enabled QoS, I set upstream to 90% of the normal speed, left downspeed on full and now massive improvement.
I get an A on the dslreports test, and I also observed e.g. if I have the tbb test running both up and down has no affect on ssh latency, smooth as silk, before it became a jerk fest.
|
|
|
|
I'm on TalkTalk Fibre.
I'm using these scripts and I'm still finding latency to be a little bit too high when downloading but not when uploading.
I read above that TalkTalk does PPP differently, does that mean I need to set a different overhead?
|
|
|
I read above that TalkTalk does PPP differently, does that mean I need to set a different overhead?
No. But I would ignore the usual TalkTalk advice of using an MTU of 1432 and use 1492 instead.
Oliver.
|
|
|
|
I'm just curious how much bandwith people here are losing with the methods to control Bufferbloat?
Is 3Mb or so a normal amount?
|
|
|
well is latency or speed capacity more important to you?
e.g. 3mbit on my downstream is about 3-4% of my max speed. It prevents packetloss and makes latency a fraction of what it would be if left alone.
For upstream 2mbit is 10%, but the effect is bigger, as upstream left alone will make latency skyrocket.
But then again I am not downloading 24/7 so I guess it isnt everything for me.
|
|
|
I'm just curious how much bandwith people here are losing with the methods to control Bufferbloat?
Since it's mainly an upstream issue, and since people rarely continuously max out their upstream for long periods of time, it's not really a huge issue.
Oliver.
|
Pages in this thread: 
Print Thread
deleted
