General Discussion
  >> Fibre Broadband


Register (or login) on our website and you will not see this ad.


  Print Thread
Standard User SpooRancher
(newbie) Tue 18-Feb-14 15:28:16
Print Post

Interpreting test results when performance is bad


[link to this post]
 
This isn't particularly tied to the ISP I'm with, but more a question on to interpret symptoms and test results....

I have an FTTC connection from Eclipse (who buy from BT Wholesale). I'm a long way from the cab, and the wiring in the area is known to be in poor condition, but for most of the time the line runs at a solid, stable 21 Mbit/s down. That's rather better than BT's speed estimator gave (17 Mbit/s)

In this state, speed tests here, and real world usage, will give me the full 21 Mbit/s for any single stream, or I'll get an aggregate 21 Mbit/s for multiple streams. Ping latency shows consistently around 18 mSec.

But the line has moments - sometimes hours, sometimes days - when throughput is decidedly bad for *single* streams. It will run badly, then it'll get better all by itself.

Testing here will give me an aggregate 21 Mbit/s for the 6 x HTTP stream test; but the single-stream TBB test will show anywhere from 2 Mbit/s to around 10 Mbit/s, and the speed can vary right across the range within a single test.

However, ping latency is still reported as around 18 mSec.

If I use BT's speed tester at http://speedtest.btwholesale.com/, I'll typically see 21 Mbit/s even when the line is running badly; but ping latency there can show 30 - 40 mSec. Only rarely will the BT speedtester show a poor performance, and the worst I've seen is 10 Mbit/s. Netstat indicates that the test is using multiple parallel streams anyway to 213.1.72.209, so wouldn't show the problem I have.

The BT speed tester isn't consistent with other testers: just now it reported 11 Mbit/s, but here gave me 19 Mbt/s and Speedtest.net gave 15.

Eclipse aren't particularly interested. When I was in this state for a few weeks they blamed it on "the Internet being slow with people downloading iOS 7". They suggested I run the detailed phase of BT's speed test, but I can't get that to work at all. Packet traces I've done of this show the PPP login works, but then no packets get a response. Eclipse have no idea what to do with the packet trace I sent them and won't help further.

The problem doesn't last long enough to get a handle on it - if I do get Eclipse to report it to BTW it'll likely be gone before they get to look, and that's £128 gone.

Does this sound to you as if there is throttling going on here (Eclipse claim they don't do any, so presumably it would be BT). Given that I can get 21 Mbit/s from multiple streams and ping latency does not get worse, am I being limited on the throughput to each individual port here?

Anyone got a helpful suggestion? What do the symptoms indicate to you?

Thanks

Alan
Standard User ian72
(knowledge is power) Tue 18-Feb-14 15:32:34
Print Post

Re: Interpreting test results when performance is bad


[re: SpooRancher] [link to this post]
 
even when the line is running badly


What do you mean by running badly? Do you mean it is performing poorly just for speedtests or are you seeing poor performance on real world applications?

My guess is that they may have an oversubscribed backhaul and you are suffering contention that will affect a single download more than a multi-stream download. Changing ISP may improve it if that is the case as a different ISP may have higher capacity pipes.
Standard User SpooRancher
(newbie) Tue 18-Feb-14 15:34:12
Print Post

Re: Interpreting test results when performance is bad


[re: ian72] [link to this post]
 
It goes badly for real world use.

Speedtest graph I've just done is here: http://www.thinkbroadband.com/speedtest/results.html...

Real world downloads show similar behaviour to the TBB test.


Register (or login) on our website and you will not see this ad.

Standard User ian72
(knowledge is power) Tue 18-Feb-14 15:36:20
Print Post

Re: Interpreting test results when performance is bad


[re: SpooRancher] [link to this post]
 
Then my guess is contention probably in the ISP network. Does this happen worst at particular times of the day? If you do a test very early in the morning (say 5AM) then do you get the same issue?
Standard User SpooRancher
(newbie) Tue 18-Feb-14 15:38:21
Print Post

Re: Interpreting test results when performance is bad


[re: SpooRancher] [link to this post]
 
For comparison, I can get weeks when the tester shows this: http://www.thinkbroadband.com/speedtest/results.html...

and real world usage gives me similar speeds.
Standard User SpooRancher
(newbie) Tue 18-Feb-14 15:39:23
Print Post

Re: Interpreting test results when performance is bad


[re: ian72] [link to this post]
 
Early mornings might be better. Today was fine to about 1100, and it's been getting worse since.
Administrator MrSaffron
(staff) Tue 18-Feb-14 15:41:32
Print Post

Re: Interpreting test results when performance is bad


[re: SpooRancher] [link to this post]
 
Suggests contention kicking in and probably on the ISP side. You can get a better for that if you have a friend on a different ISP on same cabinet and if their speeds are good then points at ISP mainly

Andrew Ferguson, andrew@thinkbroadband.com
www.thinkbroadband.com - formerly known as ADSLguide.org.uk
The author of the above post is a thinkbroadband staff member. It may not constitute an official statement on behalf of thinkbroadband.
Standard User SpooRancher
(newbie) Tue 18-Feb-14 15:47:32
Print Post

Re: Interpreting test results when performance is bad


[re: MrSaffron] [link to this post]
 
Ta - it did sound like contention to me.

Eclipse "guarantee 12 Mbit/s for the three busiest hours of the day", but they seem to be weaselling that as meaning the aggregate line speed.

May be time for a meaningful chat, if I can ever get through on the phine to them ("we are currently experiencing an unusually high volume of calls...")

Thanks, all.

Alan
Standard User SpooRancher
(newbie) Tue 18-Feb-14 17:50:26
Print Post

Re: Interpreting test results when performance is bad


[re: SpooRancher] [link to this post]
 
And here we are at 1745 back to proper performance. Gotta be contention

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

Jump to