per mac-address pcq-classifier

Sergioc70 · July 1, 2009, 3:20pm

It is currently available only ip address classification, but I need to fairly shape bridged pppoe traffic. Will a mac-address classifier be implemented?

Thanks in advance
Sergio

Sergioc70 · July 8, 2009, 9:30am

bump

thanks,
sergioc

Chupaka · July 9, 2009, 9:40am

MAC have nothing to do with IP connections. won’t

/interface bridge settings set use-ip-firewall-for-pppoe=yes

help?..

Sergioc70 · July 9, 2009, 11:11am

I need to fairly shape traffic among several pppoe tunnels but the access-point is bridging them. I won’t to be fair among multiple ip addresses my customer may have routed over a single pppoe tunnel.
So I need to fairly shape traffic among several, different, destination mac addresses.

I’m not sure about use-ip-firewall-for-pppoe, do it apply also to PCQ? Anyway it is not enough.

thaks for the answer,
ciao,
sergioc.

Chupaka · July 9, 2009, 8:29pm

no, there won’t be such functionality in PCQ… it’s Per Connection Queue. MAC is physical layer (layer 1), PCQ works on data link layer (layer 2)

Sergioc70 · August 1, 2011, 9:24am

Bump.

Maybe in one year thinhgs are changed? I still do need per mac-address classifier

Ciao,
Sergioc.

Chupaka · August 1, 2011, 10:11am

nope, PCQ still works on IP level, not MAC level, and it won’t be changed in not-so-near future

macgaiver · August 1, 2011, 10:54am

Also prioritization/limitation works trough dropping packets. It is impossible to drop encapsulated packet from within a pppoe connection without having some kind of checksum error. So there is no need for such feature cause task is impossible.

So, sorry, but you can forget about this feature request until you find some finalized RFC that describes such feature

marekm · January 22, 2019, 7:20pm

Still waiting for this feature in 2019. And yes, it is possible to drop some packets from a pppoe connection, this simply causes drop of IP packets that were encapsulated inside. Only if too many (LCP Echo request/reply) are dropped, pppoe disconnects. Long long ago, upstream bandwidth was limited (expensive) and was the bottleneck. Today, wireless is the bottleneck and can be very unfair. Why it is so difficult for the PCQ algorithm to work the same way, just looking at a different packet header field? SFQ in Linux also has gained support for external classifiers. But it needs to work on the wireless AP device, on the wireless interface queue.