Odd Fibre Speed Test result

Hi,

I have had a fibre to the home installation since Tuesday and have been generally very impressed with it. However after testing in on TBB I am a bit confused and concerned.

Any idea whats going on here?

WARNING: Your HTTPx6 speed is low compared to tbbx1. This suggests a PC or software problem.

The speed tester has warned me that my HTTP result is low and I may have a software problem, what is wrong?
Our HTTPx6 test should usually be as fast or faster than the tbbx1 test, because it uses multiple downloads to combat the effects of congestion, but in some cases such as yours and the example shown below, the HTTPx6 result is low.

The causes are sometimes your AV software that is slowing down the throughput while it inspects the traffic to ensure it is not dangerous. On some older computers or ones with limited memory we have seen the extra load of doing six tests slow things down too.

Results:

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

It suggests I have either a "Slow PC" (its a Core i5 desktop with 8Gb of memory) or my "virus software is a problem" which I disabled for the test after the first warning. There was no change.

Tried with Chrome and IE11. I am on Windows 7.

Any ideas?

The speedtester could be congested. Do other speedtests show your speed to be OK?

Yeh, other speed testers say its find but the problem seems to be that TBB says my raw download is great but my HTTP is poor. I am not sure which other testers can give me a HTTP test to check against? If it was congested wouldn't the raw throughput test show the same congestion?

http://www.speedtest.net/my-result/3593009244

Edited by deleted (Sun 29-Jun-14 12:16:17)

All speedtesters are HTTP.

The difference between the two TBB speedtests is that the 1st is a single thread and the 2nd is 6 simultaneous threads.

Obviously the 2nd increases the load by 6 times which could result in congestion.

So why does TBB state?

"Our HTTPx6 test should usually be as fast or faster than the tbbx1 test, because it uses multiple downloads to combat the effects of congestion, but in some cases such as yours and the example shown below, the HTTPx6 result is low."

No the 6 threads does NOT increase congestion it may increase the load on the PC and/or router due to the increased number of packets and processing for the test.

Because for the vast majority of people including those with faster connections this is the case.

Precise reason why yours is behaving this way I don't know at this time, once we get a better handle on who is being displayed this warning we might be able to figure it out.

Flash version can be an issue for some. Have you monitored your CPU load during the test and also tried our alternate tester at http://labs.thinkbroadband.com/ispa that does not use flash.

Well as I mentioned the PC is a fairly modern Core i5-3.6Hgz machine, with 8Gb memory with Windows 7 x64 and gigabit LAN connection to a Netgear WNDR3700v4 connected to the fibre modem.

So I am still confused why my setup would struggle with those tests?

So am I.

What firewall/AV package is running

We know the HTTPx6 can go a lot faster than this too. Will have to remember to do a grab from the database of others who got the same warning during the week and see if there is a common theme.

The CPU never goes above 55% during the entire test.

Flash test:

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

Non-Flash test:

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

Non the wiser...

Edit: Although the non-flash burst HTTPx6 outstripped the Flash burst TBB. The average was also higher and closer the to Flash TBB, but not there.

Edited by deleted (Sun 29-Jun-14 12:32:48)

The latest Avast! free version.

I tried it disabled during the tests and didn't make any difference.

In reply to a post by MrSaffron:

No the 6 threads does NOT increase congestion

It increases the load on the server 6 times.

In terms of bandwidth no it does not. In terms of CPU maybe but server is more than fast enough to serve the files multiple times to the same connection

What does the speedtest.net graph look like? You will need to do a screen shot and host it with a link.

Post deleted by Stormy

Which one?

And what about if you run a speedtest.net test on a server in London?

Like this.

http://i204.photobucket.com/albums/bb181/stormyuk/Ka...

Speedtest London:

http://www.speedtest.net/my-result/3593458699

Speetest Hull (Local):

http://www.speedtest.net/my-result/3593461257

What's your package claimed speed?

London Graph

http://i204.photobucket.com/albums/bb181/stormyuk/Ka...

The package is supposed to be "guaranteed at least 75Mbps".

Maybe traffic management is limiting you?

Maybe, I thought about that but, why it it only affecting the TBB HTTP tests and nothing else?

In the speedtest.net results it seems to activate but only down to around 120Mbps as it starts way higher. The first day I had fibre it was was 260Mbps solid.

I would say that 6 times 260Mbps would require 1.5Gbps bandwidth. Perhaps it's just not available?

Six threaded test on a 260 Mbps connection DOES NOT REQUIRE the server to have 1.5 Gbps of bandwidth available

Plus server has 10 Gbps of bandwidth to it and if got close to that alarms would go off, as 10 Gbps of traffic is pretty expensive

Its something odd with the connection, if the package purchased has a 75 Mbps guarantee with burst above then it all depends on how the bursting is detected by the ISP.

Our test and speediest.net both appear to be being impacted, e.g. one of your speediest.net graphs shows high start and then a drop away. The difference in the end figures is easily explained.

Our average is a median figure for the test.
Speedtest.net rejects 20% slowest, and 10% fastest and then picks an average (not sure if its mean or median), this often gives slightly higher figures when a test does not get a flat line.

Still not convinced.

In real world use I get around ~100Mbps (around ~13MB/s) which I am very happy with so can disregard the speed testers saying I am getting much over that.

Still does not explain the relatively low HTTP result on TBB imo. Maybe the warning is the disparity between the 200Mbps peak which is not real. Comparing ~100Mbps to ~70Mbps I get on the TBB HTTPx6 test is not a massive difference. Just odd.

The non-flash test seems a bit closer to real world on the average.

http://www.thinkbroadband.com/speedtest/button/14040...

Edited by deleted (Sun 29-Jun-14 18:07:21)

The warning is the disparity between httpx6 and the tbbx1 test, will have to pull all the offer fast KC tests and see if there is a similar pattern. If it is possible some form of traffic/burst management is the cause. Alas 200 miles away from any KC connections so difficult to sit down and debug when a provider looks to be doing something unusual

Lots of testers do not show any throughput graphs do people are none the wiser to odd/unusual behaviour. If this is normal behaviour for the faster KC lines then we can re-code and adjust the warnings accordingly.

Yeh I have asked some people with the same connections as me to run some of these tests. One guy with FTTC (Mine is FTTH) gets a graph which is more what I would expect on his 75Mbps connection.

VDSL on KC;

http://www.thinkbroadband.com/speedtest/button/14040...

I think it is the rate limiting used to cap your speed at KC that is the issue, and somehow the single thread tbbx1 is not being throttled back, or might be if you continued on for longer.

Maybe. Downloading your raw test files from;

http://www.thinkbroadband.com/download/

The 500Mb file downloaded in 26s

The 1Gb file in 43s

Both would suggest a connection in excess of 75Mbps by a fare margin. I guess I just need to enjoy the connection and not worry about it.

In reply to a post by MrSaffron:

Six threaded test on a 260 Mbps connection DOES NOT REQUIRE the server to have 1.5 Gbps of bandwidth available

How much bandwidth does it require?

260Mbps?

6 times?

Is it possible that KC see the six threads being downloaded and decide that they are trying to circumvent a restriction (of some sort) or think they are being used for P2P apps (maybe) and decides that the KC network will throttle back to the equivalent of just one ?

No, just once. A 6 thread download on a 260Mbps connection will be 6 x 43.33 Mbps.

How come it shows as 73.2Mbps?

If I knew the answer then the OPs problem would be resolved.

1. User has indicated they are on a. 75Mbps package
2. It is Fttp
3. Suggests burst rate limiting system may be issue with KC

Would need to spend a day or two in Hull to work on it. Ie take known well performing device to them

In reply to a post by MrSaffron:

1. User has indicated they are on a. 75Mbps package
2. It is Fttp
3. Suggests burst rate limiting system may be issue with KC

Would need to spend a day or two in Hull to work on it. Ie take known well performing device to them

Not quite:

In reply to a post by Stormy:

The package is supposed to be "guaranteed at least 75Mbps".

KCs Lightstream is a 50/5, 75/5 or 100/10 package so with 9.x uploads the OP should be on the 100/10 and doe that have a Min 75 guarantee?

What about the 100/30 package?

KC offer a number of packages. But seem quite conservative on the quoted figures.

50/5, 75/5, 100/10 and 350 (not sure what the upload of that is) but they gaurentee the FTTH service to be at least as good as your quoted figure upto the hard line connection.

I have seen real world performance way in excess than 75Mbps and my upload speed has always been much better than 5Mbps since install. It may be once KC has a bigger proportion of customers on Lightstream then do hard limit to 75/5 but at the moment in my experience this doesn't seem to be the case. Interestingly the first day of my install there seemed to be no trafic managment in effect as I was getting 260/20 on most testing. Unfortunately I didn't get to try the TBB speed tests on that day. It might have shed some more light on the situation.

It still might be that the x6 test is somehow confusing KC Lightstream FTTH or it is some kind of traffic management in effect.

I got a colleague of mine to run the same tests, and he see's the exact same behaviour on a 75Mbps package. Although we should remember the connection is not always limited to 75Mbps from what I can see. the TBB HTTP test is the only test I have seen which seems to have this hard stop. I will see if I can get someone on the 100/10 package to run the tests.

Still I can't complain about the perfomance of Lightstream. Finally something good to shout about from KC. Looking at the woes of BT Infinity some people have I should be happy I am served by KC.