The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2003-Jan> msg00163



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

Vpns vs explicit null label

  • From: Rahul Aggarwal <rahul@redback.com>
  • Date: Wed, 29 Jan 2003 12:03:20 -0800 (PST)
  • Cc: Mark Duffy <mduffy@quarrytech.com>, mpls@UU.NET


Hi Eric,

A couple of points on my interpretation of explicit null that has been
implemented:

1. A penultimate hop router doesn't put on an explicit null on a packet if
the outgoing label stack is not empty. Hence it has the intelligence to
insert an explicit null only if the outgoing label stack is empty. This
takes care of the VPN case, the question on which started this thread. I
think this is the right thing as explicit null is needed for payload
determination or for MPLS QoS. If there is already a label in the outgoing
label stack that should take care of it..
                                           
2. As you said the destination PE does a "fast" lookup using the fact that
an explicit null is a reserved value.

3. Destination PE rejects an explicit null if its not at the bottom of
stack. I am yet to be convinced that this is not the right thing to do as
an explicit null shouldn't show up other than at the bottom of the stack
(Point # 1). I am not sure if being liberal on what you accept applies
that well to the forwarding path. 
That said if there is a reason why this can happen, I think its not hard
to remove this restriction, unless there is hardware out there that cannot
do this.

thanks,
rahul

On Wed, 29 Jan 2003, Eric Rosen wrote:

> 
> Mark> Usually, LSRs only receive packets with labels that the LSR has itself
> Mark> distributed.  Is that true for  these labels?  Presumably not or there
> Mark> would have been no need to reserve specific values for them. 
> 
> Reserved  values were  used because  they  are easy  to look  up; it's  much
> quicker to determine that the label is  one of a small set of constants than
> it is to  look up a label in  your 20-bit lookup table.  When  a packet with
> explicit null is received, the explicit  null needs to be detected, and then
> the packet's IP  address needs to be looked up.  By  using a reserved value,
> the time  to look  up the  label plus the  IP address  becomes approximately
> equal to the time  to look up the IP address.  Without  a reserved value, it
> would be approximately double that.
> 
> Of course,  once a reserved value is  used, it does become  possible for the
> penultimate node  to put  on explicit  null even when  it hasn't  been asked
> for.  By "be liberal in what you accept", this should be handled properly at
> the egress. 
> 
> 
> 
>