OK, my plane has 3 phases,
1- testing about 20 private customers... with low transit
2- deploy small town, 130 CPE
3- deploy medium town 1200 CPE
But if accelppp- is not the GW for this transit, maybe I can continue without pci passthrough @sergey239  ?   I need for HA cluster, I think that if I enable pcipassth I will need custom migration scripts inside the cluster. But I'm not sure  what I'm saying, I will ask to my VIRT partner.

tcpdump -n -i eth0 -e -v port 67 or port 68

I'm going

I think @TomakSI and @Korvin7 have positive result with using accel-ppp on Proxmox in production

Oh, I was wrong, of course I need pass the transit by accel if I'm using IPOE, @sergey239

maybe for 1200 CPE it is ok to use bridges/vswitches on modern servers

it's not so much

At night max 3GB/s transit

We have one speedtest.net in this Cluster, that serve  8-9 Gb/s at tests , but in this VM there is pci passthrougt activated @sergey239  ;)

you will utilize cpu resources "twice" if you don't use passthrough

not really twice, but something about that

if you have modern server with a good cpu, network cards and ddr4 memory most probably it will work OK

even without passtrhough

Ok, I understand. This hosts are
32 x Intel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz (2 Sockets)

In general, you can run 2 accel-ppp simultaneously, it has load-balancing/backup mechanism for ipoe

in this case you don't need migration scripts/whatever for service backup

Ok interesting, the second accel-ppp with a delay to replies I read, is it?

please check "weight" feature in the docs

Ok, many thanks!

your cpu is an old one, I'm not sure how much traffic it can process

oky, I'm gong to do more test and diagrams, and come with news, thanks!