[Cake] feature request ;)

[Cake] feature request ;)

[Sebastian Moeller]

Hi Jonathan,

I believe we already established that cake will start out with whatever per packet overhead value the kernel already added for us (skb->pkt_len or so). Would it be possible to export the maximum of (skb->len - skb->data_len) seen over the life time of a cake instance somewhere in the statistics output? Since the kernel rules for pre-adding headers are a bit opaque this would be really helpful to select the correct per packet overhead. Say if the kernel added the ethernet header size already we do not need to use the ethernet keyword, but if it did not we should, currently the only way to figure this out is to set up an elaborate shaper and feed it with custom tailored packets and then look at the reported packet size in the qdsc statistics (if I recall correctly). It would be so much niched if cake just told us… (Case in point if I establish a shaper on eth0 the kernel helpfully adds the ethernet header, but if I use pppoe-eth0 which uses eth0 as underlayig transport the kernel does not add the ethernet header, but it MIGHT add the pppoe header size (not sure)).
	And if the max() of that field is too computationally expensive the first non-zero value would also do...


Best Regards
	Sebastian

Re: [Cake] feature request ;)

[Jonathan Morton]

[-- Attachment #1: Type: text/plain, Size: 222 bytes --]

That is an interesting idea.  But what does data_len correspond to - how
far does it unwrap the packet?  Is the difference constant between IPv4/6,
ARP, and any other protocols we might see on the wire?

- Jonathan Morton

[-- Attachment #2: Type: text/html, Size: 266 bytes --]

Re: [Cake] feature request ;)

[Sebastian Moeller]

Hi Jonathan,

On Aug 20, 2015, at 21:50 , Jonathan Morton <chromatix99@gmail.com> wrote:

> That is an interesting idea.  But what does data_len correspond to - how far does it unwrap the packet?  Is the difference constant between IPv4/6, ARP, and any other protocols we might see on the wire?

	Inquiring minds want to know exactly this, so maybe the maximum and the last seen value would be good for experimentation ;). But worst case the last seen sib->len should do as one can always try to send known size packets and the correlate the MTU with skb-len for further inferences, no?

Best Regards
	Sebastian


> 
> - Jonathan Morton