In reply to a post by CarlTSpeak:

In reply to a post by zzing123:

The only argument I can see to drop PPPoE is to provide 'greener' broadband. PPP requires an ASIC or a beefy CPU to maintain a PPP tunnel at gigabit (and beyond) speeds which is both expensive and power hungry.

So if Openreach provided a transparent L2 bridge between customer and ISP, the only CPE required is the ONT. This is easily possible nowadays and doesn't require even VPLS let alone VLANs.

The ISP in turn can turn off their very power hungry LNS concentrator, thus removing 2 boxes from the national infrastructure and saving quite a lot of power and plastic and provide their customer with whatever they want over that L2 network.

Openreach provide a layer 2 link between end user and the CP. This requires VLANs. Stacked VLANs usually, SVLAN and CVLAN. Without any VLAN usage how did you have in mind Openreach knowing how to reach each CP? Without VLANs or some other encapsulation how did you have in mind CPs separating out customers?

There is also the small matter that an ISP needs to have a link to every Openreach OLT/L2S if there's no encapsulation happening anywhere. Most ISPs can't afford this, it's time consuming, and that's why no-one sells from every CityFibre FEX. ISPs are waiting for their national product.

If you've solutions for these I'd welcome them, as your post carried a fair bit of confidence in the statements. Thanks!

https://www.juniper.net/documentation/us/en/software...

VXLAN is basically like VLAN. But instead of the 802.1q tag being limited to 4096 VLANs, you have a 24-bit VNI instead and internally you can route and manage via BGP much like a VPLS.

Edited by zzing123 (Tue 25-Jan-22 16:19:02)

I'm not against PPPoE. I'm only against making them necessary. Otherwise they're very useful. Would you mind sharing your ISP's name and router, I've never thought of *baby* jumbos?

> Name one 1gbps ISP-supplied ...

I too would be interested to see a router succeed in 900mbps PPPoE. One doesn't need to find an old router to test this, many routers allow toggling offloading. I don't think any 1Gbps router without offloading can achieve that.

Any router that allows you to set the MTU of the WAN interface and pppoe interface will allow you to do this.

I know BT support this even though they don't really advertise it. My WAN interface has an MTU of 1508 which is enough to compensate for the pppoe overhead and allows the pppoe interface to work correctly with an MTU of 1500. I know this because I tested it from a VPS in London and can see that 1500 byte ICMP packets make it through correctly:

$ ping -M do -s $((1500-28)) -c3 MY_PUBLIC_IP
PING MY_PUBLIC_IP (86.183.XXX.XX) 1472(1500) bytes of data.
1480 bytes from host86-183-XXX-XX.range86-183.btcentralplus.com (86.183.XXX.XX): icmp_seq=1 ttl=57 time=8.36 ms
1480 bytes from host86-183-XXX-XX.range86-183.btcentralplus.com (86.183.XXX.XX): icmp_seq=2 ttl=57 time=7.71 ms
1480 bytes from host86-183-XXX-XX.range86-183.btcentralplus.com (86.183.XXX.XX): icmp_seq=3 ttl=57 time=7.66 ms

--- MY_PUBLIC_IP ping statistics ---
3 packets transmitted, 3 received, 0% packet loss, time 2002ms
rtt min/avg/max/mdev = 7.663/7.909/8.358/0.317 ms

Thanks, MTU increase done. Now we need to solve the computational cost problem.

In reply to a post by weelan:

I'm not against PPPoE. I'm only against making them necessary. Otherwise they're very useful. Would you mind sharing your ISP's name and router, I've never thought of *baby* jumbos?

Baby jumbos are when you increase the ethernet frame size to carry a data payload of 1508 bytes instead of 1500 bytes.

I don't use an ISP-supplied router, I do this on my own Mikrotik routers - one FTTC via Draytek modem, one FTTP to ONT. It works on both.

The ISPs are Plusnet and Cerberus respectively.

In reply to a post by weelan:

Thanks, MTU increase done. Now we need to solve the computational cost problem.

The "computational cost" of PPPoE is so tiny you won't be able to measure it. Certainly not compared with all the other things your router is doing, especially NAT, where it has to modify every packet passing through, update the checksum, and manage state tables.

I'm convinced. It's not worth taking action to make pppoe only optional.

Aware of VXLAN and use it, it's nothing like VLAN, it's an overlay network.

I have genuinely never heard anyone suggest VXLAN as a way to build an ISP. The largest MX router Juniper have can handle 32,000 of them last I checked. You'd need tons of devices with a thousand+ VTEPs to cover all the exchanges and require a bunch of SD-WAN to steer a customer to the correct handover VTEP to deliver to their service provider.

This ignoring the issues around how to learn the hosts. Flood and learn would be interesting across a a few million VXLANs spread across thousands of VTEPs.

PPP gets shoved into L2TP tunnels to reduce 'tunnel count' across networks and aggregate them. I'm not sure how so many VXLANs, need one per end host, would work given the need to aggregate a thousand exchanges to one or two NNIs.

EDIT: While I am thinking about it you know that 802.1ad allows for stacking of more than 2 tags, right?

Edited by CarlTSpeak (Tue 25-Jan-22 19:26:52)

In reply to a post by zzing123:

In any case, you need a processor to do PPP. I'm just saying remove that entirely... Just taking BT by example: A BT Smart Hub may not be super high power at 14W. But multiply that by the 14 million customers BT have, and that's 196 MW of power being used that's unnecessary. If using FTTP, there's another 12W being used by the ONT...

Openreach ONTs typically use a fraction of that, somewhere between 1 and 3 watts.

Also don’t forget BT SmartHubs will draw more due to their WiFi radios rather than any PPPoE encapsulation overhead (which runs on hardware offload rather than the ARM cores in the SoC).