FTTPoD sigle Vs Multi thread speed issue

Hi,

I've had FTTPoD installed for a couple of months now and have an issue where my Single speed thread is about a 3rd of my multithread speed test.



I've had a ticket raised with Cerberus for a couple of weeks now but currently they can't pinpoint a reason after checking the circuit, resetting it and increasing the bandwidth available.

I've tested on multiple laptops both over my wired LAN via a router and also connected directly to the ONT. I've reset the network stacks on both laptops and checked that the network adaptor is set to 1Gbs

They have suggested an issue at my end and the next course would be an openreach visit but if nothing is found would be chargeable so I'm hoping there could be something I'm missing before going down that route.

Any ideas?

Will be nothing to do with Openreach.

Is the difference between the two speeds consistent across the time of day? i.e. if it is a lot better at night then suggests congestion/contention an issue somewhere

In reply to a post by MrSaffron:

Will be nothing to do with Openreach.

Is the difference between the two speeds consistent across the time of day? i.e. if it is a lot better at night then suggests congestion/contention an issue somewhere

Thats what I thought initially but I've tested at all times of day, evening and night. it's consistently sitting around 115 - 120. Cerberus support have also said they have ruled out contention.
I thought it may be a network adaptor issue but I've tested on different laptops, albeit they're both dell and as a wildcard, tested on my Xbox browser. same results.

Do you get the same results on Speedtest.net using the single thread option? Try a few different servers (hosts) to get confirmation. If single thread results there are ok, then nothing wrong with your connection, just that your line routing to TBB servers has issues. Btw I also never get more than 150 Mbps on the TBB tester yet nearly every Ookla speedtest server gives me > 250 Mbps on the single thread test. Not worth losing any sleep over

In reply to a post by baby_frogmella:

Do you get the same results on Speedtest.net using the single thread option? Try a few different servers (hosts) to get confirmation. If single thread results there are ok, then nothing wrong with your connection, just that your line routing to TBB servers has issues. Btw I also never get more than 150 Mbps on the TBB tester yet nearly every Ookla speedtest server gives me > 250 Mbps on the single thread test. Not worth losing any sleep over

I get similar results from speedtest, actually marginally less than TBB tester but not worth worrying about.
I've also done an 8Gb file test over FFP with Cerberus support which also gave the same output with Filezilla.
I could live with routing issues and 250Mbps.
To me this seems like a circuit issue rather than an issue with the ONT and cabling given how artificially it seems pegged around 120Mbps

These are results on my 330 Mbps FTTPoD line, though I'm with Fluidone, not Cerberus:

TBB:



speedtest.net single thread:
https://www.speedtest.net/result/8282694848.png

speedtest.net multi thread:
https://www.speedtest.net/result/8282702343.png

Might be worth asking Cerberus if they can do a Nuttcp test on your line?

That result looks rather odd.

For comparison, I'm with Cerberus Networks on SDESTBR exchange (FTTPoD 330/30 recently installed, more than one month ago).



Single Thread: https://www.speedtest.net/result/8282972079.png

- Upstream speed may seem a little odd on this test but I guess it may have been something using it on my connection at the time I did the test.

Multiple Threads: https://www.speedtest.net/result/8282974434.png

I'm using a FireBrick 2900 with a shaper applied of 267M / 24M.

Edited by Ixel (Fri 24-May-19 18:30:42)

What router are you using?

I use an ubiquiti Amplifi but have also tested with a Netgear and also directly connected to the ONT. All yield the same results.
I'm trying to eliminate any hardware issue on my side but short of the network adaptor in 2 laptops and my xbox there isn't anything else it could be. I suppose I could switch then network cable on the ONT, just in case, but lets be honest that doubtful

In reply to a post by baby_frogmella:

These are results on my 330 Mbps FTTPoD line, though I'm with Fluidone, not Cerberus:

TBB:



speedtest.net single thread:
https://www.speedtest.net/result/8282694848.png

speedtest.net multi thread:
https://www.speedtest.net/result/8282702343.png

Might be worth asking Cerberus if they can do a Nuttcp test on your line?

What would a Nuttcp test tell me? I'm not familar with this. I'll add it to my support ticket though. Can't hurt.

In reply to a post by 1nfern0:

What would a Nuttcp test tell me? I'm not familar with this. I'll add it to my support ticket though. Can't hurt.

When i had peak time slow downs on my connection last year, my ISP (Fluidone) didn't trust speedtest results from Ookla or TBB (lol), instead they ran Nuttcp tests at peak times where they could see for themselves that there were congestion issues (fixed by moving me to a different BT wholesale SVLAN at the exchange). Nuttcp involves the CP sending data packets to your pc/device to measure throughput rate, with the results plotted on a graph. No reason why Cerberus can't also use Nuttcp to measure throughput on your line.

If not already done, can you connect your pc directly to the ONT, setup a PPPoE session in Windows and then run a speedtest on the pc?

https://www.nuttcp.net

In reply to a post by baby_frogmella:

In reply to a post by 1nfern0:

What would a Nuttcp test tell me? I'm not familar with this. I'll add it to my support ticket though. Can't hurt.

When i had peak time slow downs on my connection last year, my ISP (Fluidone) didn't trust speedtest results from Ookla or TBB (lol), instead they ran Nuttcp tests at peak times where they could see for themselves that there were congestion issues (fixed by moving me to a different BT wholesale SVLAN at the exchange). Nuttcp involves the CP sending data packets to your pc/device to measure throughput rate, with the results plotted on a graph. No reason why Cerberus can't also use Nuttcp to measure throughput on your line.

If not already done, can you connect your pc directly to the ONT, setup a PPPoE session in Windows and then run a speedtest on the pc?

https://www.nuttcp.net

ok,m that seems worth taking a look at. I'm not ruling out congestion but if anything I see my speed drop a little at peak but never increasing back over 120ish after.

My tests connected directly to the ONT are using PPPoE and yield the same output which reassuring on my equipment side at least.

I was not getting the full speed. I spoke to BT FTTP support, couple days back. They said everything checked out okay at their end. They sent out the engineer to check the light level, he said it was all okay (with in spec � around -15 dB).

The speed has improved, before, It was between 130 to 270MB, now it's lot stable. I believe they must have done something at BT wholesale.

I'm using my own router Unfi USG-Pro-4.

Here is my current speed test:

<a href="https://www.speedtest.net/result/8283702682"><img src="https://www.speedtest.net/result/8283702682.png"/></a>

<a href="https://www.thinkbroadband.com/speedtest/1558737191544829855"><img alt="My Broadband Speed Test" src="https://www.thinkbroadband.com/_assets/speedtest/button/1558737191544829855.png" /></a>

Small update on my connection. Cerberus were unable to continue with escalating my issue until I had a Special Faults Investigation (SFI) visit.
Engineer arrived today and confirmed what I expected and there is no issue with the ONT. Tested again directly to the ONT with a PPPoE connection and and confirmed max 130Mbps single thread.
Hopefully the escalation can now resolve this.

Ask Cerberus to put you on a different bt wholesale svlan at the exchange. This will almost certainly fix your issue.

Thanks, I've just requested exactly that. Somewhat annoyingly, the engineer reported back to Cerberus that he could not find any speed related issue on site despite demonstrating the difference. I've had to clarify this with the support team.

G

In reply to a post by baby_frogmella:

Ask Cerberus to put you on a different bt wholesale svlan at the exchange. This will almost certainly fix your issue.

Hi, I've just had a reply back from Cerberus. The utilization of the SVLAN is only 26% with a threshold of 32.3% so my request to be moved to a different svlan was knocked back.
Before escalating further, am I missing something less obvious on my side and the problem could be with the network adaptors on my laptops restricting the single thread?
Is my expectation of single thread speeds too high and falls within acceptable range?

G

In reply to a post by 1nfern0:

My tests connected directly to the ONT are using PPPoE and yield the same output which reassuring on my equipment side at least.

In my experience all the speed test sites don't work well with anything beyond a FTTC connection. I would try downloading a Linux ISO image using either curl or wget from a range of different mirrors to see what you get. I just downloaded a CentOS 7.6 iso from clouvider.net and got around 90MB/s which on a 1Gbps connections is pretty good going.

Given you are into Ubiquiti gear if you can use WiFiman on an android device using an OTG USB ethernet adaptor.

Personally I use a server at work with a 10Gbps connection and scp as my own personal speed test server.

In reply to a post by jabuzzard:

In reply to a post by 1nfern0:

My tests connected directly to the ONT are using PPPoE and yield the same output which reassuring on my equipment side at least.

In my experience all the speed test sites don't work well with anything beyond a FTTC connection. I would try downloading a Linux ISO image using either curl or wget from a range of different mirrors to see what you get. I just downloaded a CentOS 7.6 iso from clouvider.net and got around 90MB/s which on a 1Gbps connections is pretty good going.

Given you are into Ubiquiti gear if you can use WiFiman on an android device using an OTG USB ethernet adaptor.

Personally I use a server at work with a 10Gbps connection and scp as my own personal speed test server.

Thanks, just tested the same CentOS iso from clouvider.net and getting the same output of between 16 and 17Mb/s according to chrome.so consistent with the 128-136Mbps I'm seeing elsewhere. This was tested directly to the ONT rather through the router.

Don't get me wrong, the speeds are a big step up from the 30Mbps I used to get but just trying to get a feel if this is normal and contended or artificially restricted at the exchange. I don't expect 330Mbps on a single thread.

In reply to a post by 1nfern0:

Hi, I've just had a reply back from Cerberus. The utilization of the SVLAN is only 26% with a threshold of 32.3% so my request to be moved to a different svlan was knocked back.
Before escalating further, am I missing something less obvious on my side and the problem could be with the network adaptors on my laptops restricting the single thread?
Is my expectation of single thread speeds too high and falls within acceptable range?

G

Then probably an issue on Cerberus' side, rather than BT Wholesale. If you have a smartphone with 802.11ac wifi, can you run the native speedtest app using the single thread option on it whilst standing next to your router/AP? You should be getting something like this:
https://postimg.cc/TLJZ56m1

Did you ask Cerberus if they can run a Nuttcp test on your connection?

Edited by deleted (Tue 04-Jun-19 12:17:16)

Can you try booting your laptop into Linux and downloading using that? I have seen Windows laptops which run at 100-200Mbps but can fill a gigabit when running Linux. (Usually just a case of [censored] Windows drivers provided by the manufacturer).

Or can you try with a different network adapter - preferably an Intel one? (Difficult if you're using a laptop and the ethernet adapter is built-in)

There was a separate thread where a user bought a new Intel PCI card for their desktop, and their speed improved massively. Intel just seem to make a good job of both NIC hardware and drivers.

I was of course on a Linux desktop doing that test. Though the machine a Dell Optiplex 790 is quite old now, it has been pimped over the years and sports 32GB of RAM and an SSD. However I was not using the onboard NIC, but a Broadcom NetXtreme BCM5721 out of an old server to defeat the Intel IME the motherboard comes with. Always had good results with Broadcom NIC's myself, though the 5721 does not support jumbo frames which is why it's in my desktop and not a server. Realtek is the real junk in terms of NIC's.

The onboard NIC is used for testing/setup purposes with no internet connection

Hi All,

I've been travelling a bit for work and whilst the issue was escalated with Cerberus I've avoided any testing and tweaking from my side.
The support was largely telling me the same thing with the most recent update being fairly obvious.

"Current Update: Complex fault team informed that Regarding customers query, a single thread test may not be capable of filling 330meg of bandwidth and therefor the test result may be limited to the upper level of what it can send in just one stream.
A multi-threaded test is able to send multiple streams simultaneously and is therefore able to test the full capability of the circuit.
Also they have informed us that they didnt see any issue on the connection from Network end, else the speed should vary in multi connection as well, if you run a test in different website then you should get expected speed."

That said, I began testing my connection again last night and I have seen dramatic improvements in the single thread tests. Previously they appeared capped at 130Mbps but now I'm seeing consistent results between 200-300Mbps.
There has been no changes on my side from a hardware point of view and other than perhaps a windows patch update, no software changes either. I can only think something has been changed on the their end not communicated.

Results are now

https://www.speedtest.net/result/8389391234.png

https://www.speedtest.net/result/8389389491.png


I can't really complain about this now. Of course, I don't always see the max Single thread speed but with contention and other factors, that is to be expected.