What Cake options are good for a Docsis link?

according to your experiences, which options can get good performance with cake on a docsis link

Here you go:

config queue 'eth1'
	option debug_logging '0'
	option verbosity '5'
	option linklayer 'none'
	option qdisc_advanced '1'
	option qdisc_really_really_advanced '1'
	option ingress_ecn 'ECN'
	option egress_ecn 'ECN'
	option qdisc 'cake'
	option eqdisc_opts 'nat dual-srchost ack-filter docsis noatm'
	option iqdisc_opts 'nat dual-dsthost ingress docsis noatm'
	option squash_dscp '0'
	option squash_ingress '0'
	option script 'ctinfo_4layercake.qos'
	option interface 'eth1'
	option download '312000'
	option upload '47000'
	option enabled '1'

With the obvious caveats of selecting the right (wan) port device, and appropriate download/upload limit speeds for your link...

Actually I'm not familiar with that .qos script, is that available in the stock SQM packages?

Good point, no, is not the standard script, let me simplify:

option eqdisc_opts 'nat dual-srchost ack-filter docsis noatm'
option iqdisc_opts 'nat dual-dsthost ingress docsis noatm'

I tend to be a lazy, non command line user, for this anyway, and load luci-app-sqm. And then do everything thru the luci web interface.

I typically use:
Cake
layer cake (or piece of cake)
ingress: nat dual-dsthost ingress mpu 64
egress: nat dual-srchost ack-filter mpu 64
link layer adaptation: ethernet w overhead 18 or 22

I usually use those "dangerous" catagory optimization lines, like you posted, though I tend to not use the docsis keyword, add mpu 64 there, and enter the link layer adaptation separately on the tab(ethernet with overhead, and 18 or 22 as the overhead) As I just looked up, docsis is equivalent to overhead 18 mpu 64 noatm, as a shorthand. It's in the man page for cake. So, same thing only different.

Don't know if "noatm" is needed, I just checked wondering if my not using "docsis" was leaving it out. I get it when I check with tc -s qdisc, its showing up there, even though it's not marked as a default in the man page.