Question About Per-Connection Bandwidth Limiting

[#1]

I have a question about how bandwidth is allocated on my connection.

I'm on an 80 MB/s fibre connection, and I’ve noticed that any single download stream (TCP connection) is consistently capped at around 8 MB/s, even when no other traffic is active. However, if I run multiple simultaneous downloads (e.g. 10), each one gets around 8 MB/s and together they fully utilise my 80 MB/s bandwidth.

This behavior is consistent and repeatable, and I observed the same pattern even on my previous DSL connection (with proportionally smaller per-stream throughput).

Can you confirm if this is due to deliberate per-flow rate limiting or traffic shaping implemented on your network — such as a per-session or per-connection cap? I'm not referring to my total line speed (which is fine), but rather why an individual connection can’t utilise more than 8 MB/s on its own.

I would have thought you'd allow full bandwidth to the one stream if it's the only one, and then halve it if I start another download stream so they both share the 80mB/s but that's not what's happeining.

Thanks in advance for clarifying.

[#2]

Your experience is probably caused by something other than the uplink.  These days, Plusnet does not apply any traffic shaping.

There are numerous articles describing what you are experiencing.  Do a Google search; it will deliver stuff like this - Bad throughput for single tcp connections - Server Fault

[#3]

Hmmm I'm not convinced...
Like I said, when I was on twisted pair ADSL exactly the same thing happened except the speed values were much lower, I'd be able to run 7 or 8 1mB/s streams until I was streaming my full 7mB/s bandwidth, but I couldn't have 1 stream on it's own using all the 7mB/s bandwidth

I didn't change anything in my OS, so where is the control over this happening?

This has just scaled up to 8 or 10 8mB/s streams until I'm using my full 80mB/s but I can't run just 1 download stream using all 80mB/s

So you see my 1 stream maximum jumped from 1 mB/s to 8mB/s and I dare say if I increased my line speed to 500mB/s I'd then get 50mB/s for each stream until using the full bandwidth.

So my 2 questions still apply...
1) Why can't I use the full bandwidth  (80mB/s) to download a file?
2) Why doesn't the system by default download at 80mB/s for a single download and then the more downloads added, the more it shares out the speed?
For example 2 concurrent downloads would use 40mB/s each, 4 downloads would use 20mB/s each and so on...

[#4]

So in an attempt to clarify, I have an '80' MB/s download capacity. I am reaching out to understand a specific behavior I have observed regarding my download speeds.

What I’m Observing:

• When downloading a single file, the maximum download speed consistently caps at approximately 1/10th of my total available bandwidth (around 8 MB/s), regardless of the total line capacity. (when I had 8 mB/s on the old telephone line and ADSL I got just under 1 mB/s for each download stream)

• Running multiple simultaneous downloads results in each stream being capped at roughly the same speed (~8 MB/s), and the combined throughput of these streams saturates my full bandwidth.

• This per-connection speed cap appears proportional to my total bandwidth and remains consistent even after upgrading from a slower connection (e.g., ADSL).

My Questions:

1. Does your network implement any per-connection (per-flow) bandwidth limits or traffic shaping policies that cap individual TCP connections at a fixed fraction of the total bandwidth?

2. If so, could you please clarify why the bandwidth is not dynamically shared among active connections? Specifically, why is a single download not allowed to utilize the full available bandwidth when it is the only active stream?

3. Are there options or plans to support more flexible bandwidth allocation that allows a single stream to use the full line speed when available?

Understanding these policies will help me better manage my network usage and expectations.

Thank you for your assistance. I look forward to your detailed response.

[#5]

There are a lot of answers out there if you Google the question.  For example Can you achieve full download speed on single threaded connection to a far away server instead of us...

There are a lot of complex factors within TCP/IP message handling itself, entirely not related to anything an ISP might do.  Plusnet does not profile traffic / throughput.

I would suspect that what you are seeing is a factor of ACK wait times.

@bobpullen / @dave - can you please confirm that there is indeed no Plusnet applied traffic shaping which would give rise to the reported observation?  My guess would be that the roughly 1:10 asymmetry of up and down bandwidth might be a more relevant influence.

[#6]

Thanks for the replies I really appreciate it.
But unfortunately I'm not a WWW technician, so all the three letter acronyms and technical jargon mean nothing at all to me.

I just know that a single download stream on my system no matter which OS I'm using, is somehow given roughly 1 tenth of my full potential download speed no matter what that full potential speed is.
So when I was on a 10 mB/s connection :
If I did just 1 download it only got 1 mB/s and not the available 10 mB/s
If I did 10 concurrent downloads each got 1 mB/s using up my bandwidth

Now I'm on an 80 mB/s connection:
If I do just 1 download it only got 8 mB/s and not the available 50 mB/s
If I do 10 concurrent downloads each gets 8 mB/s using up my bandwidth

I know this is not me and my OS controlling it as I use multiple OS's and they all behave exactly the same.

I just want to know where this is controlled and why the bandwidth is apportioned in this way and not SHARED over the running download streams....

Seems I can't get an answer to that.....

[#7]

ACK is an acknowledgement message which needs to be sent in response to received data.  The upstream being slower than the downstream can at times introduce delays in returning the ACK response.

Until the ACK response is received, the sender will pause.  It is not an exact nor simple science!

[#8]

Thank you for that...
Can I just ask if anyone else has the same thing happening for them?
It would really help to know if I'm alone here or if it's exactly the same for everyone.

If you have an 80 mB/s connection, and you start just one download on your system, does that download get all 80 mB/s bandwidth?

[#9]

I would suspect probably most users, which is why most of the broadband speed test tools test both single and multi-threaded.  With single thread, if there is a wait for an ACK or the need to retransmit, then the whole process stalls.  With multi threaded processing, the other threads can continue.  The Think Broadband speed test illustrates this see UK Broadband Speed Test | thinkbroadband and thinkbroadband Speed Test

My result (over WiFi) https://www.thinkbroadband.com/speedtest/1747162517127953555 and https://www.thinkbroadband.com/speedtest/1747162657596491955

It is complex and there are numerous potential bottlenecks in each component of the end to end connection from your device to the remote server.

[#10]

If I go to Microsoft.com and download a windows 11 ISO file, I get 67Mbps download and 2.2Mbps upload. That is on a Full Fibre 74 connection - I do have "Quality of service" enabled on my ER605 router which limits maximum throughput a little but improves bufferbloat scores.

Which router are you using?

[#11]

The other posts cover the points I would make, but there is a sharing mechanism, if you can fully utilise the Internet connection.

With one device (on a wired connection) run a speedtest (say Ookla) and see what it says.

On the multiple thread test it will probably fully use both download and upload.

Then with two devices (both on wired connections) run two tests at the same time, the achieved speed will probably be roughly equally split between the two devices.

It will be less than the single device test achieves.

It's your devices, not anything Plusnet is doing to your connection.

[#12]

I tried my suggested test (Plusnet 150Mb/s fibre) - One device

One speed test

Two devices (both cabled to the router)

Two speed tests

There is a practical difficulty in that I couldn't quite trigger both tests at the same moment so there is slight "flattery" on the two device test but it's clearly less than a single device and does show the connection being shared between the two devices.

[#13 · Accepted solution]

OK, this is gonna take a lot of doing guys, but I know exactly what's happening.....

Yours truly is confusing MegaBYTES per second with MegaBITS per second.

I feel like such a complete fool, and I'm so sorry to have wasted all your time.

I really appreciate all your help....

On my old ADSL line I was only ever seeing the maximum download speed of around 1 MBYTE/S
Which equals 8 MBITS/s which was my ADSL line speed...

Now I'm on the slowest fibre connection which is around 74 MegaBITS per second so the fastest download I'm seeing is around 9.25 MegaBYTES per second and that's what I'm seeing.....

So sorry to have made a fool of myself and wasted all your time!

I'll get my coat! 

[#14]

Dave,

Do not be so hard on yourself, you’re not the first and you will not be the last to have made this error.

I did write under about the tool you were using for assessing the speed: proper speed tests report in mbps, where as data transfer tests often report in mBps.

I did nearly ask the question but you seemed confident with your reported results … which to some extent fitted feasible expectations.