I am pleased to see this basic improvement to qos-scripts. I will be testing this patch soon in the cerowrt-1.0rc7 series if it doesn't get committed to openwrt head soon. I've had very puzzling results with the prior version that perhaps this will clean up (see http://www.bufferbloat.net/issues/259 as one example)
I note that qos-scripts has other problems than these. 1) It does not account for IPv6 at all. that could be handled 1b) There are also issues with encapsulated packets, (6in4, gre), as only TOS information and skb-priority survive the transition to encapsulation. I'm not sure how to address that in any way. 2) Things like openssh over ipv6 are not setting the TCLASS field in the first place... (https://lists.bufferbloat.net/pipermail/bloat-devel/2011-August/000215.html - fix is described at: http://www.bufferbloat.net/issues/249 . Nearly no application is doing TOS-like things over ipv6 - dropbear, etc - and assuming the code described on 249 is right, it's trivial fixes across the applications to do more of the right thing here) 3) It is very rare that RED kicks in and starts ecn-ing packets in the default qos-scripts 4) Newer algorithms such as DRR and SFB should be more 'pluggable' somehow. I'd like actually to be able to emulate 'famous' shapers such as wondershaper and adsl-optimizer in some way. 5) It seems like a good idea to also do qos on internal interfaces at this point. 6) there is noooooo 6. 7) A currently unsolvable problem with the present day network stack architecture is measuring the weight of multicast packets, which can be 300x+ more slow to transmit than normal packets, over wireless. 8) A second unsolvable problem is the interaction of all shaping techniques with packet aggregation over wireless-n. On Fri, Sep 23, 2011 at 4:37 PM, Weedy <weedy2...@gmail.com> wrote: > On 16/11/10 01:55 PM, Ben Pfountz wrote: >> This patch updates qos-scripts to support fair traffic sharing using the >> SFQ with external classifiers method. It also corrects a bug in the >> unsupported ESFQ method already used by qos-scripts. (ESFQ: >> http://fatooh.org/esfq-2.6/ only updated to 2.6.24, it was switched to >> an SFQ patch after that and not updated since 2008) >> >> A class can be forced to use SFQ, and an external classifier added like >> this: >> >> config class "Normal" >> option avgrate 10 >> option priority 30 >> option packetdelay 100 >> option limitrate 94 >> # option qdisc "sfq perturb 2" >> config class "Normal_up" >> # option filter "protocol all flow hash keys src divisor 1024" >> config class "Normal_down" >> # option filter "protocol all flow hash keys dst divisor 1024" >> >> Using these options, the user needs to load cls_flow before qos-scripts >> starts. >> >> I've got more information here: >> http://oneitguy.com/blogs/netprince/fair-traffic-sharing-esfq-broken-switching-sfqexternal-classifiers >> >> >> This has been tested on r23914. >> >> Signed-off-by: Ben Pfountz <netprince<>vt_edu> > > I am bumping this so we can get a current patch. > _______________________________________________ > openwrt-devel mailing list > openwrt-devel@lists.openwrt.org > https://lists.openwrt.org/mailman/listinfo/openwrt-devel > -- Dave Täht SKYPE: davetaht US Tel: 1-239-829-5608 http://the-edge.blogspot.com _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
