[Cake] squash option

[Cake] squash option

[Dave Taht]

Inbound diffserv markings cannot be trusted. A simple remark to
dscp & ECT(3) of diffserv is kind of a must for cake.

I note that my original patch to do this probably wont work due to the
COW nature of the skb at that point.

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

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

Re: [Cake] squash option

[Jonathan Morton]

> On 13 Apr, 2015, at 01:20, Dave Taht <dave.taht@gmail.com> wrote:
> 
> Inbound diffserv markings cannot be trusted. A simple remark to
> dscp & ECT(3) of diffserv is kind of a must for cake.

As I noted previously, cake can trivially be configured to ignore Diffserv.  A simple iptables rule would then suffice to remove the mark for downstream consumers if desired.

More useful would be the ability to re-mark packets according to local Diffserv policy before cake’s classifier acts on it.  Iptables doesn’t help here, because it runs *after* ingress qdiscs (even in ifb).  Unfortunately, the likely complexity of the configuration mechanism for such a device runs counter to cake’s deliberately simple configuration interface.

However, I have no objection to a second qdisc which could be run in series with cake to perform that function.  This would be a qdisc which performs no queueing of its own, and simply performs the re-marking before passing the packet up the qdisc stack.  The configuration mechanism of this can be relatively complex, but narrowly focused on the re-marking task.  Call it “icing” or “cherry” or whatever, to emphasise that it goes on top of cake.

 - Jonathan Morton

Re: [Cake] squash option

[Dave Taht]

On Sun, Apr 12, 2015 at 5:01 PM, Jonathan Morton <chromatix99@gmail.com> wrote:
>
>> On 13 Apr, 2015, at 01:20, Dave Taht <dave.taht@gmail.com> wrote:
>>
>> Inbound diffserv markings cannot be trusted. A simple remark to
>> dscp & ECT(3) of diffserv is kind of a must for cake.
>
> As I noted previously, cake can trivially be configured to ignore Diffserv.  A simple iptables rule would then suffice to remove the mark for downstream consumers if desired.

We need the simplicity of an rfc-compliant squash option in cake.

FQ is far more important than diffserv or aqm in sorting things
properly for most people. As noted elsewhere, what we see are packets
marked CS1 entering the system (treated as priority by most ethernet
devices) and then treated as background on wifi devices.

> More useful would be the ability to re-mark packets according to local Diffserv policy before cake’s classifier acts on it.  Iptables doesn’t help here, because it runs *after* ingress qdiscs (even in ifb).  Unfortunately, the likely complexity of the configuration mechanism for such a device runs counter to cake’s deliberately simple configuration interface.

I concur that remarking is too complex for cake. But squashing is not.
The ip header is hot in cache at the time at which it could occur,
also.

> However, I have no objection to a second qdisc which could be run in series with cake to perform that function.  This would be a qdisc which performs no queueing of its own, and simply performs the re-marking before passing the packet up the qdisc stack.  The configuration mechanism of this can be relatively complex, but narrowly focused on the re-marking task.  Call it “icing” or “cherry” or whatever, to emphasise that it goes on top of cake.
>
>  - Jonathan Morton
>



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

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