I managed to connect the Internet with my 7362SL Router two ways. As Client behind a 7560 working as Modem and the 7362SL beeing Modem itself with PPPoE. I had to extract the Modem FW from an 7340 AVM Firmware.
Connection is shown with 116 MBit but only performs 60Mbit.
Both ways, as client and as modem.
Upload is fine with 40Mbit/s
With stock Firmware the 7362SL is performing with nearly 100Mbit, so i guess it is no hardware issue.
It's not quite clear if you're referring to the established link speed (modem side) or the effectively usable data rate reported by speed tests. The lantiq SoC is not the fastest, without assistance it tops out somewhere between 55-60 MBit/s, you do need to enable software flow-offloading to push its performance just beyond 100 MBit/s, it's not fast enough to reach those speeds on its own.
If you simply bride the DSL with one of the LAN ports and don't do anything else computationally intensive (like WiFi) an xrx200 (BT homehub 5a) can transfer an aggregate 124.6 Mbps goodput over a gross sync 116.8/37 Mbps link with a theoretical upper limit of
116.8 * 64/65 * ((1500-8-20-20-12)/(1500+26)) = 108.52190745
37 * 64/65 * ((1500-8-20-20-12)/(1500+26)) = 34.3776590382
108.5+34.4 = 142.9 Mbps
So bidirectional saturation of a 100/40 link seems slightly out of reach, but in aggregate an xrx200 can reach reliably >> 100 Mbps... without flow-offloading but also without routing, fire-walling or other gunk, ad one CPU core is pretty much maxed out during that test.
My understanding of the OP infers that the 7362sl is used as router, terminating the PPPoE session in both scenarios ("As Client behind a '7560 working as Modem'" and "the 7362SL beeing Modem itself with PPPoE") - and that's (routing, NAT, firewall and PPPoE, which -by itself- is already on the heavy side) where the lantiq SOC is on its knees.
Performance may also differ between devices using the same SOC, e.g. while I could get full speed on a bthub5 (modem + router/ PPPoE (and AP) with software flow-offloading on a 100/40 MBit/s VDSL2+vectoring line), I did not manage to nudge an Easybox 904xDSL beyond 55-60 MBit/s (regardless of software flow-offloading or disabling FXS/ enabling SMP). I have no idea why, but it simply didn't break the 60 MBit/s barrier.
Oh, I agree xrx200 routers will not be able to do much useful at 100/40 dsl link speed, so IMHO the only viable way forward is to use such a device as bridged modem and combine it with another device acting as router (performing all the typical chores, like NAT/firewall, PPPoE, WiFi). The "advantage" would be to allow optimal positioning for the modem (as close to the telco socket/TAE/APL as possible) and for the WiFi AP(s).
IMHO the amazing thing is how resilient these ~10 year old MIPS based router designs around the xrx200 proved to be, these came out when DSL speeds topped-out at around 50 Mbps for many networks and before vectoring and G.INP became wide-spread, but they also have pretty much reached close to the end of their usefulness as AIO devices... (I think non-OpenWrt proprietary OS heavily leaned into accelerators to make such devices last a bit longer).
The recent performance improvements in master, speeding up the dsa drivers, are also impressive (won't help modem usage or PPPoE much, only for ethernet-to-ethernet routing).
Yes, it looks like xrx200 is getting a lot of love in its sunset days No complaints from my side, I will happily take all these improvements (according to my tests, bidirectional DSL link saturations seems limited not by the ISP shaper or the DSL sync, but by SIRQs maxing out the one CPU handling the interrupts, so anything making DSL or Ethernet more efficient should help in getting closer to maxing out the link). If I was adventurous I would try to move either the DSL or ethernet interrupts to the second CPU core, but I think I read of crashes others experienced when trying that. Also my understanding is that these are more like SMT siblings and not like fully independent cores so I am not sure whether moving interrupts around will help much, still tempting though.
thanks again for your feedback. @slh advice to enable the software offload pushed the speed to 108Mbit/s
i am honest: i am not able to follow your disscussion in depth, because my understandig is not that far for now.
my point is i am not satisfied with my fritzbox 7560 when it comes to managing, because it is damn slow at the ui (around 50 devices in the homelab). internetspeed and reliability are not a problem.
i thought about pfsense (which means to get new hardware) and edgerouter x (which is not available) to improve my expirience.
after some research i found that my 7362SL (i have 3 of them and not all are needed) are compatible with openwrt, so i thought to give it a try.
even if it is off topic. what would be your suggestion? get some more powerfull hardware for openwrt? or use 2 7362SL and splitting the tasks (in which way?).
one óf the next steps will be splitting the homelab, so that IOT devices will use an other VLAN than the rest.
No complaints from my side, I will happily take all these improvements (according to my tests, bidirectional DSL link saturations seems limited not by the ISP shaper or the DSL sync, but by SIRQs maxing out the one CPU handling the interrupts, so anything making DSL or Ethernet more efficient should help in getting closer to maxing out the link). If I was adventurous I would try to move either the DSL or ethernet interrupts to the second CPU core, but I think I read of crashes others experienced when trying that. Also my understanding is that these are more like SMT siblings and not like fully independent cores so I am not sure whether moving interrupts around will help much, still tempting though.