[Cake] cake integration woes

[Cake] cake integration woes

[Dave Taht]

I finally had a chance to read over the various bits of code flying in
loose formation.

-1) If you have any mods to cake left to push please push them so I
can get into various builds for the weekend.

0) The main iproute2-cake repo needs to be updated to mainline head or
at least have one patch integrated (which I just did for the
ceropackages build, which does not have everything else integrated yet
(see 5)).

Still don't know how to push an update to it. Why not just use github?

1) the tc code for cake has "diffserv" only which I think defaults to
the diffserv4 setting. I will add a diffserv8 to the command line
options.

2) It is really not clear to me how the relative order of priorities
works on rate unlimited.

3) I really want that squash option. I like being able to build this
thing out of tree, so I can fix and debug easily (and fix the cow
issue also). Am sufficiently lazy to not do it if -1 exists....

4) I don't like the precedence option being in there at all.

5) Somewhere I seem to have misplaced the tc code for fq_pie...

6) Has the testbed been run yet?

-- 
Dave Täht
Open Networking needs **Open Source Hardware**

https://plus.google.com/u/0/+EricRaymond/posts/JqxCe2pFr67

Re: [Cake] cake integration woes

[Jonathan Morton]

[-- Attachment #1: Type: text/plain, Size: 796 bytes --]

I've been concentrating on writing material for the paper, so there isn't
any new code right now. There's several things that I'm fairly confident I
know how to implement, but not actually done yet.

If "git push" doesn't work, talk to Toke. It does for me.

In "unlimited" mode, the priority DRR never flips from the
priority-balanced quanta to the bandwidth-balanced ones. This is an
extension of what happens when the shaped rate is above the physical link
rate. Starvation of lower priorities still doesn't happen, but you don't
get the soft bandwidth limits.

The cake3 iproute2 module I pushed originally had separate Diffserv 8 and 4
support, with "diffserv" being a synonym for the newer 4-class one. If the
code renaming has gone and lost that, that's not *my* fault.

- Jonathan Morton

[-- Attachment #2: Type: text/html, Size: 951 bytes --]

Re: [Cake] cake integration woes

[Jonathan Morton]

> On 30 Apr, 2015, at 19:25, Dave Taht <dave.taht@gmail.com> wrote:
> 
> 3) I really want that squash option.

And I really don’t think it belongs in cake.  Not unless we put a complete DSCP re-marking system in there, which would massively complicate the configuration interface, which we don’t want to do.

Let me emphasise: cake can already *ignore* DSCP.  So can anything else in a network, with the possible exception of stupid wifi interfaces, which in the long run we should try to fix.  If you’re using Diffserv internally, then you already have a re-marking system separate from cake, which can do the squashing for you.

And, on a more technical note, adding squashing means cake crosses a threshold - from not modifying packets (except for ECN marking, which has a special function for the job) to doing so.  I’d rather have a separate, specialised qdisc for the re-marking task, and put squashing in there instead.

> 4) I don't like the precedence option being in there at all.

This is perhaps fairer comment, since I don’t expect many people will use it.  But since it’s already in there, and isn’t very much code, it might not make sense to remove it.

 - Jonathan Morton