PPPoE offload, bridge mode

Wondering if it's possible to use a Mikrotik device that has PPPoE offloading, in transparent, or bridge mode to handle just the PPPoE connection, but not do any routing, firewall, etc?
I have a network set up and working great, except my gateway can't handle my 3gb/s fiber connection since it uses PPPoE. When I test with the ISP provided router, I get full speed, but it has a built-in chipset for handling PPPoE hardware offload.

Thanks!

I think at the moment no MikroTik device exists that support hardware offloaded PPPoE. And this is not possible either:

Routing is always needed because the PPPoE interface is presented as a Layer 3 interface only. You can "split" the PPPoE processing to the ISP router that has hardward offload for PPPoE, let it handle PPPoE dial-out and put the MikroTik router behind it (WAN of the MikroTik router in the ISP router's LAN) and configure DMZ on the ISP router (MikroTik router's IP address as DMZ address).

In case you want IPv6 and the ISP gives you a large subnet, you might need to create multiple DHCPv6 clients on the MikroTik router, each with a different DUID, to be able to use multiple /64 subnets, because I think most ISP modem routers only handout /64 prefixes to their LAN clients (when DHCPv6 is used).

The following seems to work, but unlikely at 3Gb/s

https://forum.mikrotik.com/t/using-rb5009-in-bridge-mode/175205/8

It looks like the RB5009 can do exactly what I want, but is limited to around 1.5gbps.
Is there a newer product that has a more powerful CPU, or can handle multi-threaded PPPoE?

Not really. Implementation of PPPoE is exactly the same on all Mikrotik devices (i.e. single-threaded and no HW-offload) so the throughput of PPPoE depends (almost) exclusively on single-core CPU performance. And there RB5009 is at the top of device roster.

If you refer to the PPPoE performance, not including potential fancy complex layer 7 filtering, then this is not true. Even without fasttrack enabled the RB5009 can achieve > 2.4 Gbps PPPoE throughput (that's the limit that I could test, because I am limited by GPON), with one or multiple PPPoE client instances.

Ok good to know! 2.4 is a lot better than what I can get right now.
Are you using your RB5009 in this transparent PPPoE setup?
If so, would you be willing to share your config?

It's with PPPoE on the RB5009, but the RB5009 also acts as normal router, not the setup from the post mentioned above. I only have WAN through PPPoE. Like I wrote, 2.4 Gbps is probably not the real limit because the router is only at 70% CPU load without fasttrack when doing that. And here is an older post with the screenshot with the CPU load WITH fasttrack:

You can see that TX and RX on ether1 and sfp-sfpplus1 (where the GPON stick is plugged in) reached 2.4 Gbps while CPU usage is below 50% with fasttrack. So the limit is from my ISP and GPON. GPON also limits the upload at 1.2Gbps and my ISP only allows 1 Gbps upload.

As for the config: It actually just the defconf as starting point, I even used QuickSet when it first popped up to switch from DHCP to PPPoE. Since then I've added more rules to the defconf firewall (but the accept established / related rule that handle most packets when fasttrack is not active is at the same position anyway, and the added rules are below it so have no big effect).

That setup achieved the speed above. I've since made it even more complicated by moving all ports into the bridge, even the GPON port and use VLAN-only. The PPPoE now sits over a VLAN interface under the bridge, that has the sfp-sfpplus1 port as access port, so in theory even more overhead. The screenshot above is from this setup.

Also, for a long time RouterOS had no fasttrack for IPv6 and speedtests done with IPv6 have no problem reaching that load. The reported throughput in the speedtest app is a bit lower (a few tens Mbps lower) due to the IPv6 header overhead, but the rates on the ethernet ports are the same. Now RouterOS has fasttrack for IPv6 too and it works well.

So PPPoE does not overload the RB5009 at 2.5 Gbps. However, looking at the linked post. If you applied that setup with mark-routing then you'll need to disable fasttrack. So the CPU load will probably raise to 70% for 2.5 Gbps.