Maximum achievable speed on 500/75 FTTP (Openreach)?

This is purely out of interest, I am happy with my FTTP connection.

I am with TalkTalk (residential) and I am on the 500/75 package. Speed tests from various places put my download speed at a consistent 515 Mbps. My upload is at a consistent 72 Mbps.

My question is, is this a cap set by Openreach? Do they actually have a higher cap than 500? What about the upload, do they actually cap below 75?

Openreach have 17 speed tiers for FTTP, but ISPs typically adopt only a subset. There are 3 higher tiers than what you have currently (as defined in the Supplier Information Note here on pages 3/4). For instance TalkTalk only offer one speed tier higher than you currently have. Upload limits are defined in the same way. There are higher and lower limits depending on the product tier.

Operationally the ISP own systems then shape / rate limit the speeds to meet the requirements of the nominal Openreach speed tier. Openreach systems police this to make sure that the actual traffic rates meets the product rate.

The wholesale product from Openreach is 550M down, 75M up. That's the rate at which the OLT shapes the packets in and out.

The values you measure with a "speedtest" measure the TCP payload only. There are TCP headers, IP headers, possibly PPPoE headers (although not with TTR), and ethernet headers on top of that. So the actual payload throughput you see is lower than the headline figure - roughly 95% (e.g. 1456 / 1524 but the exact value depends on a number of things)

The ISP *may* perform their own shaping of download traffic at the BRAS as well, and if they do it will be set slightly slower than the OLT rate. This can be good if the OLT is doing aggressive packet dropping, and should give you a better overall experience. I don't know if TTR does this.

In reply to a post by candlerb:

The wholesale product from Openreach is 550M down, 75M up. That's the rate at which the OLT shapes the packets in and out.

The values you measure with a "speedtest" measure the TCP payload only. There are TCP headers, IP headers, possibly PPPoE headers (although not with TTR), and ethernet headers on top of that. So the actual payload throughput you see is lower than the headline figure - roughly 95% (e.g. 1456 / 1524 but the exact value depends on a number of things)

The ISP *may* perform their own shaping of download traffic at the BRAS as well, and if they do it will be set slightly slower than the OLT rate. This can be good if the OLT is doing aggressive packet dropping, and should give you a better overall experience. I don't know if TTR does this.

Thanks, so on that basis about 72 Mb looks almost spot on the money for 95%, so it looks like TalkTalk probably cap at about or just below the Openreach rate?

And it looks similar on the download, 516 is just below the 95%, so again it looks like TalkTalk probably cap at about or just below the Openreach rate?

Do other ISPs apply a more aggressive cap? I wonder if this is to manage bufferbloat?

In reply to a post by Gigabit:

In reply to a post by candlerb:

The wholesale product from Openreach is 550M down, 75M up. That's the rate at which the OLT shapes the packets in and out.

The values you measure with a "speedtest" measure the TCP payload only. There are TCP headers, IP headers, possibly PPPoE headers (although not with TTR), and ethernet headers on top of that. So the actual payload throughput you see is lower than the headline figure - roughly 95% (e.g. 1456 / 1524 but the exact value depends on a number of things)

The ISP *may* perform their own shaping of download traffic at the BRAS as well, and if they do it will be set slightly slower than the OLT rate. This can be good if the OLT is doing aggressive packet dropping, and should give you a better overall experience. I don't know if TTR does this.

Thanks, so on that basis about 72 Mb looks almost spot on the money for 95%, so it looks like TalkTalk probably cap at about or just below the Openreach rate?

And it looks similar on the download, 516 is just below the 95%, so again it looks like TalkTalk probably cap at about or just below the Openreach rate?

Do other ISPs apply a more aggressive cap? I wonder if this is to manage bufferbloat?

The rate is a sync* rate, there is overhead so you will get a little below the quoted rate, regardless of provider (unless the provider is over-provisioning, which on Openreach they won't be because that's not a thing).

You'll get a little below the sync* rate, and your tests suggest you're about spot on.

* as someone has pointed out, sync rate is the wrong terminology here, but the point is, the ONT will be at 550/75, the throughput you can have is a little less in practical terms. Either way, your results look good.

Edited by therioman (Thu 18-Aug-22 11:15:00)

In reply to a post by therioman:

The rate is a sync rate, there is overhead so you will get a little below the quoted rate, regardless of provider (unless the provider is over-provisioning, which on Openreach they won't be because that's not a thing).

You'll get a little below the sync rate, and your tests suggest you're about spot on.

The rate on FTTP isn't a sync rate. It's your share of a larger pipe and is measured at Ethernet layer. The ONT always syncs at the full rate of the PON and the rate quoted doesn't include the GPON overheads.

I used to get closer to 80Mbps on the upstream, but looking at test history the past week or more i'm only getting 74 as a max. Not sure if something has changed.

As a comparison, I'm on Cerberus "300/50", where the underlying Openreach product is 330/50.

On wired speedtest I get consistently 306M down, 51.5M up. This suggests to me that I'm being BRAS-shaped by Cerberus downstream, and that Openreach slightly over-provisions the upstream.

That's with with local shaping turned off. Normally I also have a bit of bufferbloat tweaking enabled on my Mikrotik:

Text
1 23 4	/queue type add kind=sfq name=sfq-default sfq-perturb=10/queue simple add dst=pppoe-out2 max-limit=50M/330M name=nuc1 queue=sfq-default/default target=10.0.0.0/8,2001:db8::/56

With this in place, I still get 306M down but upload hits about 48.5M, since it's being rate-limited on my own side.

The benefit is that interactive sessions like ssh are able to "queue jump" for lower latency while large uploads are taking place. That is: you can trade a little peak performance for a better overall experience. The BRAS shaping is probably doing the same for me in the download direction.

In reply to a post by Gigabit:

Thanks, so on that basis about 72 Mb looks almost spot on the money for 95%, so it looks like TalkTalk probably cap at about or just below the Openreach rate?

And it looks similar on the download, 516 is just below the 95%, so again it looks like TalkTalk probably cap at about or just below the Openreach rate?

Do other ISPs apply a more aggressive cap? I wonder if this is to manage bufferbloat?

Openreach actually cap slightly above the upstream product rate.
With most providers you will see just over 75Mb/s throughput on the 550/75 package.

Talktalk seem to cap upstream throughput a few Mb/s below what it needs to be but it isn't the end of the world.

I'm on Vodafone 550/75Mbps Tier (via Openreach) and the actual speed is pretty constant at 520/70Mbps.