Use RPS or RFS https://www.kernel.org/doc/html/latest/networking/scaling.html?highlight=rps

this is not related to accel, because accel-ppp is only control-plane for pppoe sessions so slow performance of pppoe is related to kernel, not to accel-ppp

This question is not related to accel, traffic forwarding is done by kernel, not by accel-ppp so please contact netdev (https://lore.kernel.org/netdev/ ) if you want to discuss it

It is not fixed in actual version. It is required to fix here: https://github.com/accel-ppp/accel-ppp/blob/2b865db72bc2ddc6411950d72f1c23e8ef115b8a/accel-pppd/ppp/ipcp_opt_ipaddr.c#L100 , https://github.com/accel-ppp/accel-ppp/blob/2b865db72bc2ddc6411950d72f1c23e8ef115b8a/accel-pppd/ppp/ipv6cp_opt_intfid.c#L209 , https://github.com/accel-ppp/accel-ppp/blob/2b865db72bc2ddc6411950d72f1c23e8ef115b8a/accel-pppd/ctrl/ipoe/ipoe.c#L668

also, we have to check driver performance when ipoe_lookup_hwaddr is used for many sessions (many sessions in vrf)

It will not work if there in the conditions below:

1. two sessions in L2 ipoe shared mode
2. the same interface (e.g. eth0)
3. first session without vrf
4. second session with vrf
5. the same ip.src for both sessions

I use this one: show sessions match calling-sid vlan123

why do you want to have "uplink interface" as a separate column? why calling-sid/called-sid with interface name is not enough?

@micron
regarding to interface:

ifname-in-sid=called-sid|calling-sid|both:

You can include "uplinkif" to calling-sid or called-sid, it is already implemented. Please check man/doc related to this feature

please check this patch https://github.com/accel-ppp/accel-ppp/pull/31

please enable max debug level and provide debug log retaled to the session from the beginning to the end

Proposed solution doesn't solve described issue because 'show loaded-config' prints last loaded config disregarding what is applied. Even if accel doesn't apply an option from loaded config, it will be printed by 'show loaded-config' command

please try this branch https://github.com/svlobanov/accel-ppp/tree/show-loaded-config

@themiron please review PR https://github.com/accel-ppp/accel-ppp/pull/24

@ProLan please try to apply patch below:

I can confirm that linux kernel learns from arp.src.hw_mac, not from eth.src. It means that current behaviour for proxy_arp=2 is useless. If no l2 isolation, then proxy_arp is not required. in case of isolation, linux clients will not be able not communicate to each other

It might be an issue with vlan_mon, but ipoe module will work with multiple accel instances

you can simply run multiple instances of accel-ppp

Your accel-pppd version is too old (142c943721615020bca80de4c69e6bbf574529aa = Mon Oct 22 12:00:02 2018 +0200)

You are talking about ifcfg,mode and start options, right? These are per-interface configuration, not per-session so I could not imagine how to configure per-interface options via access-accept message.

Can you explain your idea in details? What do you want to add? Which messages should be modified? (access-request/access-accept/accountint-request?)