On Fri, Jan 27, 2006 at 03:34:48PM -0800, Caitlin Bestler wrote:
> [EMAIL PROTECTED] wrote:
> > On Fri, 2006-01-27 at 15:32, Roland Dreier wrote:
> >>     Roland> No, I think trying to create a mapping is a bad idea. 
> >>     The Roland> semantics of VLANs and IB partitions are sufficiently
> >>     Roland> different that it's probably better to treat each
> >> concept     Roland> natively. 
> >> 
> >>     Steve> Roland, can you expand on this some?
> >> 
> >> I don't think transport neutral code should be dealing with either
> >> P_Keys or VLANs.  The Linux model for handling VLANs is that each
> >> VLAN has a separate network interface.  So an iWARP consumer should
> >> never deal with VLANs, just with a routing choice of interfaces.
> >> 
> >> Similarly if a consumer is using the iWARP-emulation CM for IB, then
> >> the P_Key will come from the IPoIB interface.  Only native IB
> >> consumers that understand partitions ever have to deal with P_Keys.
> > 
> > What about a gateway between iWARP and IB ? Would it need the
> > mappings between VLAN and IB partition ? If so, I would
> > presume that is at a layer above what you are talking about.
> > 
> 
> If you are attempting to implement an iWARP/IB gateway *above*
> transport neutral verbs then I don't think there is anything
> that can be defined that will be of much help.

The IB side of the gateway should know about P_Keys, but I think
attempting to translate that to vlans is a bad idea. Any use case I can
think of that is going to want to map IB partitions to a vlan will do it
by having the gateway know what P_Key goes to what ethernet interface,
and there are already other tools to map ethernet interfaces to vlans.
Let's not reinvent this until there's an actual need.
_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general

To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general

Reply via email to