Re: [lwip-devel] Bridge Implementation

[Joel Cunningham]

> On Feb 10, 2017, at 12:54 AM, Simon Goldschmidt <address@hidden> wrote:
> 
> Joel Cunningham wrote:
>> What performance issue did you uncover?
> 
> Nothing to uncover, I just think there's too much of the generic networking 
> code involved. It takes too much CPU cycles. That might be OK if you need 
> some of the fancy features or want to go from briding to routing. But for a 
> simple "switch" replacement, it seems oversized.
> 
>> Most stacks seem to have a bridge concept that is at least defined in man 
>> pages/online documentation,
>> not sure I’ve seen the concept defined in an IEEE document.
> 
> Ehrm, isn't that what 802.1D is about?

Thanks, I’ll review that again.  If I remember, that doesn’t deal with the 
complications of having the bridge also run a layer 3 host

> 
>> We would have to define a set of features for a bridge in LwIP
> 
> I'm not sure that really fits into (core) lwIP. lwIP is more about upper 
> layers. Although it would be nice to somehow have multi-port support in lwIP 
> (e.g. to implement an LLDP receiver on a multi-port device), the bridge seems 
> more like a netif thing (one IP address but multiple ports).
> 

Agree that the bulk of the bridge implementation would probably be some kind of 
virtual netif, but the one piece that I need to think/learn more about is how 
ARPing for the bridge’s IP address works because there is a different hardware 
address for each interface

Joel