FTTH 1000/300 Mbps Sky MAP-T with Keenetic HERO AX
Honestly, I am incredulous of the performance (in IPv4 with offloading), I am used to seeing poor results on most of the hardware I have seen/tested.
I want to congratulate all the Keenetic staff who did a great job.
In particular a raspberry pi only works with a customised image made by a forum.fibra.click user that modifies the load balancing management between the cpu's for a moment and requires a specific NIC tp link.
Despite this, very good performance is only achieved in IPv6, a speedtest server in IPv4 witout offloading gives results in line with or worse than other devices with OpenWrt, but with offloading the performance are good!. In any case, the values eventually settle at very similar devices such as the belkin ax3200 with a very similar hw.
3.8.5 (no offloading)
3.9 Beta 1 (offloading)
Then we come to the MTU side, which has never been written about on this forum but I imagine the engineers who have worked on it know: MAP-T requires an extra 20 bytes of MTU per packet in IPv4, there are two solutions:
increase the IPv6 MTU to 1520
decrease the IPv4 MTU to 1480
If not handled correctly this causes problems on sites such as atm.it, ebay.
All of these sites do, so again I congratulate the person who made this implementation. What's more, it works the site that never worked in the openwrt implementation of MAP-T www1.sky.com/opensourcesoftware/
As you can see from speedguide.net/analyzer.php it is reduced to 1480 because MTUs above 1500 are not supported, the other option would also be convenient.
what is missing from this implementation?
implementation of the possibility of requesting an IPv6 prefix (sky leaves the possibility of requesting a prefix that you have already obtained via ia_pd), this was implemented by a user of the forum.fibre.click on openwrt, the patch can be found here https://github.com/edofullin/odhcp6c
Jumbo Frame to allow you to have a MTU > 1500 and thus not have to have a 1480 MTU in IPv4. (the OP tells me it is possible via CLI, as soon as I have some time I will test it)
Better IPv4 performance, performance in IPv6 is very good (I hope that improving the hw/sw offloading will bring about the canonical 930 Mbps in IPv4).