The MPLS WG Archive

[David Charlap <David.Charlap@marconi.com>]

Hitesh Kulkarni wrote:
> 
> In a few current or next gen boxes hw genarally has capability to look 
> into the payload even for labelled traffic for load distribution 
> purposes on aggr links etc right??

Was there a question in here?  Asking a working group about whether 
somebody has implemented something in hardware isn't exactly a useful 
question.  You may be better off asking groups that deal with users of 
protocols, or developers of hardware.

I think you meant to ask whether hadware with this capability may use it 
for forwarding labeled traffic.

My answer is that it can use whatever it wants, but it would be a dumb 
idea.  Unless you want your hardware to snoop RSVP LABEL_REQUEST objects 
and also monitor for Martini-type labels in the stack, _and_ have 
configured knowledge about any exceptions, your hardware won't always 
guess correctly about what the LSP's payload is.  If it's not the type 
you think it is, then you will be reading garbage values, and you'll be 
making incorrect decisions based on that garbage.

If you think a packet carries an IP payload when it's really AppleTalk, 
or IPX, or Ethernet frames, or raw voice data, or something else, and 
you start doing load-balancing forwarding calculations based on what you 
think you find in there, you're going to break the traffic.

-- David