Thanks, appreciate it.

In my experience, limited single-threaded throughput is due to low-level packet loss.

TCP treats packet loss as a signal of congestion, and slows down to avoid the network collapse - as it is designed to do.

However, packet loss which is *not* due to congestion will also cause TCP to slow down. Examples include links with low level bit errors, or microbursts of packets overflowing the limited buffer space in some switch in the path.

The formula is here:
https://en.wikipedia.org/wiki/TCP_tuning#Packet_loss

The maximum throughput of a single TCP connection, in bytes per second, is MSS / (RTT * sqrt(Ploss))

If a single stream is limited to 300Mbps with 9ms round-trip time, this formula predicts your packet loss is

(1460/.009/(300000000/8))^2 = .0000187

which is about 1 in 50,000 packets dropped; a very tiny amount of loss indeed.

The TCP throughput you can get goes down as the RTT goes up. This becomes a real problem when doing inter-continental transfers.

Such tiny amounts of packet loss are very hard for the ISP to diagnose, especially as there are all sorts of places this might occur, and it's understandable that the ISP says "there's no problem".

However, iDNet do risk losing their reputation as "top class" service provider if they can't get to the bottom of it, when other ISPs demonstrably perform better. If they're serious about it, they could install perfSonar nodes at various points in their network and measure packet loss between them.

If you really need to upload or download large files at such high rates, the best way to do it is with multiple TCP streams, using a protocol designed for this application, e.g. GridFTP, or even Bittorrent.

Edited by candlerb (Tue 05-Aug-25 14:53:35)

I was hoping you wouldnt ask as I am pretty sure I closed the tab now, I will see if I can find it in my history.

Had a look and there is way too much in my history, I did searches to see if I could weed it out, but nothing sorry.
Bear in mind it could be a partner issue, it could be pppoe, the thread was mainly the user trying to figure things out, then near the end someone blamed idnet as it all started when they moved from one isp to another.

I remember it also saying the issue went away with windows 11, windows 11 has a network stack that resists packet reordering, as well as latest versions of linux. So if you already using windows 11 its not the same problem.

Edited by Chrysalis (Tue 05-Aug-25 16:14:34)

4:30PM Seems to be much worse now:

Connecting to host speedtest.idnet.net, port 5201
[ 5] local 192.168.1.63 port 51078 connected to 212.69.36.111 port 5201
[ ID] Interval Transfer Bitrate
[ 5] 0.00-1.00 sec 48.5 MBytes 405 Mbits/sec
[ 5] 1.00-2.00 sec 36.0 MBytes 302 Mbits/sec
[ 5] 2.00-3.00 sec 44.0 MBytes 370 Mbits/sec
[ 5] 3.00-4.00 sec 26.5 MBytes 221 Mbits/sec
[ 5] 4.00-5.01 sec 46.5 MBytes 389 Mbits/sec
[ 5] 5.01-6.01 sec 41.9 MBytes 351 Mbits/sec
[ 5] 6.01-7.01 sec 44.8 MBytes 374 Mbits/sec
[ 5] 7.01-8.01 sec 49.0 MBytes 412 Mbits/sec
[ 5] 8.01-9.01 sec 46.9 MBytes 392 Mbits/sec
[ 5] 9.01-10.00 sec 33.2 MBytes 282 Mbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate
[ 5] 0.00-10.00 sec 417 MBytes 350 Mbits/sec sender
[ 5] 0.00-10.00 sec 416 MBytes 348 Mbits/sec receiver

In reply to a post by jalzoo:

In reply to a post by Croftie2:

Thanks, that looks a lot better than mine, That gives me hope it can be improved somehow. I'd be currious how it looks in the evening. Do you know if your backhaul is cityfibre or zen?

Im on cityfibre backhaul that goes to Telecity. I believe zen backhaul shows redbus in traceroutes

Ill do some more perfs later during peak hours.

I'm fairly sure I'm on Zen backhaul as I'm never affected when CityFibre National has issues, yet a traceroute looks like this:

traceroute to bbc.co.uk (151.101.128.81), 32 hops max, 3 probe packets per hop, 84 byte packets
1 212.69.63.54 <telehouse-gw10.idnet.net> 3.554 ms 3.418 ms 3.456 ms
2 212.69.63.136 <telehouse-gw8.idnet.net> 3.560 ms 3.666 ms 3.623 ms
3 212.69.63.169 <telehouse-gw7.idnet.net> 3.656 ms 3.733 ms 3.535 ms
4 * * *
5 151.101.128.81 <bbc.co.uk> 3.957 ms 3.868 ms 3.847 ms

So I don't really know.

I'm fairly sure I'm on Zen backhaul as I'm never affected when CityFibre National has issues, yet a traceroute looks like this:

That would be Cityfibre National backhaul that you are on.

With zen backhaul your first hop would be something like this: redbus-gw11.idnet.net

If CityFibre's business services have a different backhaul then that could explain things, but we've not been hit by any of the National outages.

I can conform on AA via CF I do get 900+ on TBB's upload test.

In reply to a post by Chrysalis:

I can conform on AA via CF I do get 900+ on TBB's upload test.

I was with AAISP with cityfibre but i changed due to the problems they were having with their routers firmwares. Was getting lots of drops in connection, Have they sorted that out now?. If they have i might have to buy out my contract with IDnet and change back to them.

No problem thanks for looking. On W10 but will try live booting linux.