Kitz ADSL Broadband Information
adsl spacer  
Support this site
Home Broadband ISPs Tech Routers Wiki Forum
 
     
   Compare ISP   Rate your ISP
   Glossary   Glossary
 
Please login or register.

Login with username, password and session length
Advanced search  

News:

Author Topic: General upstream slowness  (Read 2549 times)

Weaver

  • Senior Kitizen
  • ******
  • Posts: 11459
  • Retd s/w dev; A&A; 4x7km ADSL2 lines; Firebrick
General upstream slowness
« on: April 12, 2021, 04:11:02 PM »

The problem I have now is that all known speed testers are reporting a low effective TCP upstream speed: I’ve tried speedtest.aa.net, testmy.net/upload and the Ookla speed tester speedtest.net and between them they report 0.29 - 0.6 Mbps, whereas ~1Mbps is expected.

Modem 3 is currently turned off because of the packet loss problems that I have been having, and these measurements were made with that modem off, and my expectations are adjusted accordingly.

Live sync rates:
  #1: downstream 2593 kbps, upstream 666 kbps
  #2: downstream 3052 kbps, upstream 534 kbps
  #3: off
  #4: downstream 3247 kbps, upstream 582 kbps

I’m wondering if the slowness could just possibly be due to the exceptionally high rate of modem 1, sounds very unlikely but that’s the only obvious thing that has changed.

Efficiency: sync rates 666+534+582 = 1782 ; adjust for protocol efficiency: 1782 * 0.88 = 1568 expected IP rate, which needs to be slightly downgraded further to allow for IPvx and TCP overheads. So we are a long way short of the expected TCP combined upload speed.

Logged

Weaver

  • Senior Kitizen
  • ******
  • Posts: 11459
  • Retd s/w dev; A&A; 4x7km ADSL2 lines; Firebrick
Re: General upstream slowness
« Reply #1 on: April 12, 2021, 06:15:15 PM »

Talked to AA about this. They couldn’t see anything that I had not spotted. As an experiment, they have set all lines to 446 kbps upstream but the effective TCP upload speed is still really slow.

This rules out firebrick vs tcp problems, due to variation in speed between lines. Also the possibility of a problem with line 3 is ruled out, because that modem was turned off.

When the lines are back to their normal, varied speeds, I am going to do individual per-line upload tests for AA.
Logged

Weaver

  • Senior Kitizen
  • ******
  • Posts: 11459
  • Retd s/w dev; A&A; 4x7km ADSL2 lines; Firebrick
Re: General upstream slowness
« Reply #2 on: April 13, 2021, 02:00:12 AM »

AA forgot to remove the upstream speed caps on the lines which they had imposed at the DSLAM by magic, or they intended to leave it this way. So I haven’t done individual speed tests in case this is not how they wanted it, it’s not ‘realistic’. I’ll check with them what they want before I get into it.
Logged

Weaver

  • Senior Kitizen
  • ******
  • Posts: 11459
  • Retd s/w dev; A&A; 4x7km ADSL2 lines; Firebrick
Re: General upstream slowness
« Reply #3 on: April 13, 2021, 03:56:50 AM »

With caps in place, in the wee small hours I’m getting 7.07 / 0.60 TCP rate from the AA speedtester with three modems only (modem 3 off). With the Ookla speed tester (website speedtest.net but I’m using the iPadOS app) I am getting 7.0 / 0.87. The Ookla speedtester is pretty good as far as these things go. With testmy.net/upload I get 540 kbps upstream. That last tester is very convenient because you can do uploads only, so it saves time and download bytes, and you can choose the size of the file it uploads. It seems to me to be a very straightforward TCP thing: generate a file n bytes long, transfer it, time the transfer, and that’s it. I can’t know for sure what’s going on with it but that’s how it seems to work.

Several tests performed with each tester.
Logged

Weaver

  • Senior Kitizen
  • ******
  • Posts: 11459
  • Retd s/w dev; A&A; 4x7km ADSL2 lines; Firebrick
Re: General upstream slowness
« Reply #4 on: April 13, 2021, 12:56:23 PM »

I refer back to earlier threads including this one https://forum.kitz.co.uk/index.php/topic,25690.msg431868.html#msg431868

There have always been unexplained fast seasons and slow seasons regarding measured combined upstream, even though sync rates do not change. Perhaps this is just a very much more extreme example, a 50% drop from 1.0Mbps, like a rise from 1.0Mbps to 1.65 Mbps according to speedtest.aa.net.uk given in that earlier post. Even though in that thread, that particular speed tester was described by some posters as junk, it doesn’t explain things this time round, because the speed test results are roughly consistent across three different speed testers: the AA one, the testmy.net/upload and the Ookla app.
Logged

Weaver

  • Senior Kitizen
  • ******
  • Posts: 11459
  • Retd s/w dev; A&A; 4x7km ADSL2 lines; Firebrick
Re: General upstream slowness
« Reply #5 on: April 13, 2021, 04:58:44 PM »

Individual-line tests (downstream / upstream); all at 446k capped upstream sync as seen by FB in PPP connect message .

line 1 only
AA speedtester : 2.33 / 0 Mbps !! upstream test failed, for some reason
testmy.net: 289 kbps upstream
ookla : 2.22 / 0.29 Mbps

line 2 only
AA : 2.35 Mbps / 0.0 !! upstream test failed
testmy : 296 kbps upstream
ookla: 2.34 / 0.33 Mbps

line 4 only
AA : 2.61 / 0 !! upstream test failed
testmy : 250 kbps
Ookla : 2.52 / 0.28 Mbps


Spreadsheet below, also attached is (i) a zipfile containing an iOS Numbers app format spreadsheet and (ii) attachment that is an xlsx zipfile.

Line   1   2   4   
sync rate   0.446   0.446   0.446   
observed TCP   0.29   0.33   0.28   
expected IP    0.375   0.375   0.375   
observed/expected IP   0.774   0.881   0.747   should be nearly 1.0, but not quite because of IP+TCP headers
modem loading factor    0.950         
protocol efficiency    0.8844         



So the FB is currently only uploading stuff at 75 - 88% of the expected speed although need to give it an extra 2% or so to allow for IP and TCP headers, converting from IP PDUs to TCP SDUs (ie TCP payload).
« Last Edit: April 13, 2021, 06:43:52 PM by Weaver »
Logged

Weaver

  • Senior Kitizen
  • ******
  • Posts: 11459
  • Retd s/w dev; A&A; 4x7km ADSL2 lines; Firebrick
Re: General upstream slowness
« Reply #6 on: April 16, 2021, 02:53:30 AM »

I just did an overall combined upstream speed test with testmy.net/upload. What is so great about this tool is the simple straightforward nature of its design (apparently); it seems to be just a simple TCP transfer, but also you can choose the size of the test file it sends in either direction. I chose a much upload file this time, 12 MB and then 25MB. On one of the 12 MB uploads it scored an 890kbps upload which is nearly twice the speed of a lot of earlier 400-500k tests. It seems that unsurprisingly a larger transfer gives a more reliable result, but before the tests done were not bogus because the results were fairly consistent across multiple speed testers; roughly speaking, within the noise present generally in each one tester. I’m now using these large test uploads to generate enough traffic so that I can see the whole thing clearly in the AA CQM clueless graphs because their duration is now long enough to show up properly.
Logged