Congestion Isn't Just VM Only

[Ignitionnet]

With the exception of those on Talk Talk, Sky and some Zen customers, all fibre services at the moment use BT Wholesale. BT Wholesale have some congestion issues on their network, in some cases fixable via configuration changes but in some cases requiring new hardware and fibre.

Think Broadband Story
RevK of Andrews and Arnold
AAISP Congestion Report
Plusnet Congestion Report

These issues are somewhat inevitable given increased service speeds, but for BTWholesale are disappointing as these guys make their money selling bandwidth to ISPs, so it's pretty ridiculous them not upgrading their network to keep up with demand from their customers given they're paying per Mbit.

The indignation from Neil McRae was a bit of a PR disaster too.

[qasdfdsaq]

Refresh my memory here - don't ISPs buy "pipes" of some fixed bandwidth to exchanges/offload points whatever they're called, and historically it's been these that have become full as a result of the ISP not buying enough capacity rather than the underlying infrastructure itself filling up?

If so, how does what we're seeing now differ? So far all the above links just point to monitoring consumer endpoints, with nobody seeming to look at any internal endpoints within the BTw network, such as an old TBB monitor I'd set up ages ago (before TBB stopped me being able to set up/modify my monitors) to a BT Retail RAS

[Ignitionnet]

Originally Posted by qasdfdsaq

Refresh my memory here - don't ISPs buy "pipes" of some fixed bandwidth to exchanges/offload points whatever they're called, and historically it's been these that have become full as a result of the ISP not buying enough capacity rather than the underlying infrastructure itself filling up?

If so, how does what we're seeing now differ? So far all the above links just point to monitoring consumer endpoints, with nobody seeming to look at any internal endpoints within the BTw network, such as an old TBB monitor I'd set up ages ago (before TBB stopped me being able to set up/modify my monitors) to a BT Retail RAS

An ISP controls their bandwidth, they know if they've capacity issues and they know which customers map to which endpoints on their network. If they see some customers having strife while others aren't on the same MSILs they can drill down, as Plusnet and AAISP do, to exchange level.

They can also go to BRAS level as they know which BT Wholesale BRAS customers connect via.

From that lot they can tell, if so minded, whether the issues are on an entire BRAS, on a geographically close group of customers but on different exchanges, pointing to an issue at a metro node or BRAS LAG, or at an individual exchange, if some customers at an exchange an SVLAN congested, if all customers at an exchange showing issues potentially the backhaul the SVLANs live on is full.

The BTW 21CN network has individual SVLANs mapped to BRAS via metro nodes, so congestion can occur at logical SVLAN level, or at a physical level in a couple of places.

EDIT: Customer <xDSL> Exchange <SVLAN on LAG/GbE/10GbE> BRAS <Ethernet> LAC <Ethernet> ISP LNS

[Mr Angry]

The last time I went to BRAS level I got my face slapped!

[qasdfdsaq]

Originally Posted by Ignitionnet

EDIT: Customer <xDSL> Exchange <SVLAN on LAG/GbE/10GbE> BRAS <Ethernet> LAC <Ethernet> ISP LNS

Thanks, that's what I was looking for!

[Ignitionnet]

Originally Posted by Mr Angry

The last time I went to BRAS level I got my face slapped!

You're doing it wrong.

I was about to make an exceedingly tasteless joke but will get my backside kicked and am trying to behave.

[Sirius]

Originally Posted by Mr Angry

The last time I went to BRAS level I got my face slapped!

[Ignitionnet]

I PM'd the tasteless joke, Mr A.

Seeing a minimal amount of congestion here; barely perceptible but due to the usually excellent average latency it does show as an increase along with a touch of packet loss.

I have FTTC from Plusnet and BT Retail, so both use BT Wholesale's network and I know that's the common point between them.

[rhyds]

Do you think that BT/The LLU Cos/VM have managed to build up enough resiliency in their core/backhaul networks to cater for the widespread use of 50+Mbit broadband? Once the prices start to fall I can see a fair few problems emerging

[Ignitionnet]

Originally Posted by rhyds

Do you think that BT/The LLU Cos/VM have managed to build up enough resiliency in their core/backhaul networks to cater for the widespread use of 50+Mbit broadband? Once the prices start to fall I can see a fair few problems emerging

For now. On the whole usage per customer is still less than 1Mb/s. Well engineered networks alongside the wider availability and lower cost of 100GbE ports make a difference.

You can shove 20 of these into a single MX2020 chassis.

[Ignitionnet]

Eep, just actually tested connections rather than a brief look at the TBB meter...

[qasdfdsaq]

Y'know last time I suffered congestion on BT (temporary for about 3 days) it was bad enough to send my download speed down to 4Mbps from 76Mbps yet didn't have any noticeable increase in minimum or average latency, with only the max latency jumping by about 10ms.

[Ignitionnet]

Originally Posted by qasdfdsaq

Y'know last time I suffered congestion on BT (temporary for about 3 days) it was bad enough to send my download speed down to 4Mbps from 76Mbps yet didn't have any noticeable increase in minimum or average latency, with only the max latency jumping by about 10ms.

Makes perfect sense. The way DOCSIS and the Ethernet network where this is happening on BT handle congestion are quite different. Downstream congestion on VM looks kinda similar to the congestion on BT, obviously upstream congestion is completely different as you're comparing a full-duplex transmission medium against a TDMA request-grant medium with buffering done on individual CPEs.

Here's what bad 21CN congestion looks like.

[craigj2k12]

Isnt there also the fact that BT use QoS where VM dont?

[Ignitionnet]

Not on 21CN; that applies to 20CN, the old Colossus ATM network.