Lower Single Thread download speed on toob full-fibre

Greetings to all from here in Gosport Hampshire. I've just had toob 900 Mbps up and down installed. However, I'm getting lower single thread download speeds as seen on links below.... I've bypassed the Linksys MX400 series router provided and connected directly with a single ethernet cable between my main PC and their modem black box, for these speed tests.... Any advice or help welcome please... Many thanks in advance, Mark.

https://www.thinkbroadband.com/speedtest/17001659957...

Single Thread (below)

https://www.speedtest.net/result/15517229366

Multi-thread (below)

https://www.speedtest.net/result/15517223718

One of the issues with high speed fibre is finding a speedtest that is able to meet your speeds.

Have you tried the Fast.com one which is pretty basic but generally keeps up?

From work - https://www.speedtest.net/result/15519985636

Edited by smouty (Fri 17-Nov-23 13:27:25)

Lower single-threaded performance usually means some low-level packet loss is taking place. If this is consistent (i.e. not just at peak hours) then it's probably a bad switch or router setup in Toob's network - Google for "microburst". If it only happens at peak hours, then it's likely congestion.

You can get a better proof of this by doing a Speedtest to a site in Europe, and to one in the USA, and see if you get progressively lower speeds. This is because the impact on TCP throughput is a factor of both packet loss and round-trip time. The formula is here:
https://en.wikipedia.org/wiki/TCP_tuning#Packet_loss

Example: a RTT of 10ms and packet loss of 0.01% (just 1 in 10,000 packets!) would limit a single-threaded TCP stream to

1460 / (0.010 * sqrt(0.0001)) * 8 = 117 Mbps

I saw this with an Easynet leased line connection once. After proving the issue to them, they were able to identify a bad switch port, and after moving me to a new port, the problem was gone.

You may have difficulty persuading Toob to look into this though. They may just say "everything is fine, 1 in 10,000 packets lost is not a problem". And it could of course be a problem at your side, e.g. a dodgy NIC, so you want to rule that out first by testing with a completely different client device.

In reply to a post by mrmarktigger:

Any advice or help welcome please...

Hopefully you've tried a few ethernet cables. If using Windows have you bypassed/uninstalled any third party security software (e.g. Kaspersky, ESET, AVG etc) as most of these products are not tested with high speed network connections sadly. As an alternative you could boot into a live Linux environment to test the hardware. Also could be faulty network card in your PC, or drivers.

Lastly, have you asked Toob ?

I've seen this with both zen and you fibre on fttp using tbb. Even on my 2gb/2gb you fibre single thread is not great. I wonder if its a tbb thing. You fibre tested my link with a hardware 10gb device and the results were bang on.

In reply to a post by mrmarktigger:

Greetings to all from here in Gosport Hampshire.

There are people in Southampton reporting similar on the ISPreview forum for toob.

In reply to a post by candlerb:

......
Lower single-threaded performance usually means some low-level packet loss is taking place. .
......

Many thanks to you all for your input.... It has made me look into this issue a bit deeper... Just as a bit of background this end, everything on my home network is working fine. I'm able to transfer files at 900 Mbps plus on my home network. Including my 3x Raspberry Pi's...

@candlerb I've been investigating packet loss...

So I started watching some youtube clips too... I've done some ping tests and trace route tests, and, others too... However, Nothing showed up for packet losses on those, Until.....

Along this line I found and Australian guy recommending a GUI tool for latency / packet loss....

Sure enough it shows up in excess of 25% packet loss at times on two nodes /routes to this and most sites in the UK. I'll try and figure out how to post the screen shots these results very soon....

In reply to a post by jchamier:

In reply to a post by mrmarktigger:

Any advice or help welcome please...

Hopefully you've tried a few ethernet cables. If using Windows have you bypassed/uninstalled any third party security software (e.g. Kaspersky, ESET, AVG etc) as most of these products are not tested with high speed network connections sadly. As an alternative you could boot into a live Linux environment to test the hardware. Also could be faulty network card in your PC, or drivers.

Lastly, have you asked Toob ?

Many thanks @jchamier, I can recall you from past being at extremely helpful with your past experience and helping out many.... What I have seen from an app so far is:


ae-7.r20.londen12.uk.bb.gin.ntt.net 129.250.4.140

and

ae-13.a03.londen12.uk.bb.gin.ntt.net 129.250.3.249

Both have in excess of 25% packet loss at times (off peek)

Many thanks to you all

How are you establishing that these have 25% packet loss on traffic throughput?

If you're reading results from pinging or tracerouting through these nodes, all you know is that 25% of that type of traffic (e.g. ICMP echo for ping) directed at that node is being dropped. This is a fairly normal rate-limiting configuration to prevent a router being overloaded by responding to these requests which could compromise the ability to perform it's core function.

Packets dropped on pinging a router node may indicate possible congestion or packet loss on that link or node, but abolutely doesn't prove it.

FWIW, I've struggled to get consistent good speed tests from TBB on toob, whilst others elsewhere, and running iperf tests on high capacity links under my control haven't indicated a more general problem. I do feel like there's some odd latency and my gut feel is that the IPv6 config on the supplied router is perhaps partly to blame (things are anecdotally better on v4 only).

From what I've seen in the networks I manage, it almost feels like one of the peering links between TBBs upstream and toob's peering is a LAG of lower speed connections rather than a single higher speed interface, which could present as lower single thread speeds depending on the configuration

In reply to a post by mrmarktigger:

Along this line I found and Australian guy recommending a GUI tool for latency / packet loss....

Sure enough it shows up in excess of 25% packet loss at times on two nodes /routes to this and most sites in the UK. I'll try and figure out how to post the screen shots these results very soon....

That's junk. Please forget what you saw: bin it along with Covid deniers. If you had 25% packet loss then your throughput would be limited to <1Mbps.

Seeing dropped packets in the middle of a traceroute is *not* evidence of packet loss. It's just that routers deprioritise and/or rate-limit their responses to traceroute probes.

To get the behaviour you observe, you would have *very low level* packet loss. To detect this directly would require something fairly sophisticated like Perfsonar (which by default sends 10 packets per second) running over an hour. And you'd either have to run your own remote Perfsonar node, or find a public one which accepts tests from commercial IP addresses (most only accept from R&E address space)

Instead, I suggest again that you do the following tests:
- single threaded to a server in UK
- single threaded to a server in Europe
- single threaded to a server in USA
(speedtest.net lets you select the target server). If these show speeds which reduce in proportion to the RTT (ping time), then this is very strong evidence for low-level packet loss, and you can calculate it from the formula given.