Pedro A. Aranda writes:
> I don't know, since I'm quite new to the inner world of OpenSolaris. But from
> my (perhaps) naïve point of view, I see two situations.
>  
> 1) Label Edge Routers: it's were the 'tunnels' are created.
> Here is where we have to know _what_ we want the FEC to
> mean, i.e. we have to trigger processes for traffic enginnering
> related stuff here.
> When you want to transport IP packets on top of LSP, I really see a 
> possibility of triggering the process via dladm. I have to confess that I 
> have no clue as to what one should do if we wanted to transport 
> Ethernet frames over it. 

I think you'd likely be bridging in that case, and these would still
look like links to dladm.  That doesn't seem special.

The special case is when you're using MPLS for TE.  The FEC in this
case selects both destination *and* class of traffic.  That doesn't
really look like any kind of "interface" to me, so representing it as
a tunnel in dladm seems wrong.

I suppose I could see other solutions here.  If the system were
redesigned so that interfaces could have duplicate addresses and
overlapping subnets that differ only in traffic class, then
representing LSPs as tunnels all the time would make sense.  That
seems to me to be a much more radical change than just allowing
special FECs to "cherry pick" some traffic to follow a special path.

> 2)  Label Switch Routers: this is the point were all the magic of pushing, 
> popping
> etc. is done without necessarily knowing what that should be good for and 
> what traffic
> over the LSP is being transported. They have to run the TE and non-TE routing 
> protocols.

LERs must know how to push and pop as well.  They're just not doing
any push-AND-pop (swap) action between LSPs.

LERs also have to participate in TE behavior, depending on exactly
what the deployment is doing with MPLS.  In particular, if you're
using RSVP-TE, the LER is the one who starts the PATH message.

> I really see two different projects here. And if I have to decide what should 
> be started
> first, then I would go for the Label Edge Router first, making sure it 
> interworks with 
> commercial LSR platforms and only afterwards would go for the Label Switch 
> Router
> project.

I'd suggest otherwise.  I think it's far too easy to design an
LER-only implementation that doesn't extend well to produce an LSR.
I'd rather start with a core design for MPLS (plumbing, ILM, label
handling) and produce something on top of that (perhaps LER-only at
first), than to design something specifically for the edge case.

-- 
James Carlson, KISS Network                    <[EMAIL PROTECTED]>
Sun Microsystems / 1 Network Drive         71.232W   Vox +1 781 442 2084
MS UBUR02-212 / Burlington MA 01803-2757   42.496N   Fax +1 781 442 1677
_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to