Thanks, now that I see them, I agree they are weird, especially since the CSs that are slower than the others are not really static, but they change between tests...
Did you know that if you direct password-less ssh access to the router you can use the "--remote-metadata=root@192.168.1.1" command line argument to flent to record the qdisc setup of your cake box (assuming 192.168.1.1 would be the cake box's address in your local network, that way it gets easier to check what options where in use for each interface. It will also save an equivalent of "tc-s qdisc" on that remote metadata before and after the flent test ans store it in the flent result file. But unless you set up passwordless logins via sharing the ssh keys, this will not work well.
Best Regards
Hi moeller0,
thanks for the tip, I'll try it for the next tests.
Some tests from the last week shows a different result: it seems setting ECN also on the ingress isn't ignored. https://we.tl/tZ9NnLgSp8
I'm planning to re-run all the tests straight from the client to a netperf server with the debloater in between, this way I'll avoid possible interference with the ISP and/or my local network (mis)configuration.
Lorenzo
Argh, right you are, cake will always honor ECN, but without an ingress cake instance this will have no influence on the traffic passed. You can disable ECN at your endpoint to get nonECN behaviour from cake. So in your case ingress ECN setting should be immaterial as you only have egress cake instances (but these will honor ECN).
Great, that should give good data...
Hi all,
here is the configuration I'm using since a couple of weeks:
/etc/config/sqm
config queue 'eth1'
option interface 'eth0.1'
option qdisc 'cake'
option script 'layer_cake.qos'
option debug_logging '0'
option verbosity '5'
option qdisc_advanced '1'
option qdisc_really_really_advanced '1'
option linklayer 'none'
option squash_ingress '0'
option squash_dscp '0'
option ingress_ecn 'NOECN'
option egress_ecn 'ECN'
option download '0'
option upload '15360'
option eqdisc_opts 'dual-dsthost diffserv3 wash'
option enabled '1'
config queue
option interface 'eth0.2'
option qdisc 'cake'
option script 'layer_cake.qos'
option debug_logging '0'
option verbosity '5'
option qdisc_advanced '1'
option qdisc_really_really_advanced '1'
option linklayer 'none'
option squash_ingress '0'
option squash_dscp '0'
option ingress_ecn 'NOECN'
option egress_ecn 'ECN'
option download '0'
option upload '15360'
option eqdisc_opts 'dual-srchost diffserv3 wash'
option enabled '1'
bufferbloat is greatly reduced and fair sharing between hosts is working well too!
1 Like