380.67 Beta 3 is now available, with a short, but important changelog.
Code:
04eec3c Bumped revision to beta 3
2e39e39 ctf: revert BCM6.37's ctf to pre-7743 version, as an attempt to fix broken PPPoE acceleration
23f8e63 Merge with GPL 7743 binary blobs for RT-AC87U; updated kludges
d992e24 qos: implement overhead support for AdaptiveQoS; re-design overhead configuration on the webui; add non-atm based overhead support
927d443 rc: duplicate ssl_enable keyword when enabling FTP TLS
51256fc Add space before WAN IP
b99dc76 qos: implement wedge to iproute2 to insert (fq_)codel support in Adaptive QoS
Beta 3 adds support for fq_codel to Adaptive QoS. This is done by intercepting calls made by the Trend Micro component to the Linux traffic classifier, and replacing sfq settings on-the-fly with fq_codel settings. The same technique is also used to implement support for configurable overhead values for Adaptive QoS.
The overhead configuration was also overhauled for both Traditional and Adaptive QoS. Now, you can enable or disable ATM support (previously, anything >0 was automatically set to ATM). The value can also be manually entered, or chosen from a list of preset configurations. I'm unsure of the actual impact of this parameter, as the documentation on it is rather sparse, even conflicting at time. My best recommendation is that you experiment with them. If in doubt, keep things at the default value (0, with ATM disabled).
The screenshot I posted earlier show the latency of a ping being sent every second to
www.google.ca while running a speedtest from speedtest.net. As you can see, fq_codel does help in keeping average latency lower (aside from a few occasional spikes, yet those aren't as high).
My thanks to the group of volunteers that confirmed that the patch system was working properly. Do note that fq_codel is not a magic bullet. It won't make your connection go faster, and it might only help in some specific scenarios. You might also need to experiment a bit with the overhead value. But in general, it should be an improvement in terms of latency when having multiple concurrent streams.
Finally, in an attempt to resolve the lack of NAT acceleration for PPPoE for some models, the NAT acceleration component was reverted to an older version. Please confirm whether this makes any difference if you have an AC56/AC68/AC87 and a fast PPPoE connection that had performance issues with beta 1 and 2.
