[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PWE3] draft-bryant-pwe3-packet-pw-02.txt

HENDERICKX Wim wrote:

Luca, for IP you are right in terms of protocol stack but I have the
impression that the packet-PW draft is more generic and we should be
open minded to other applications apart from IP.

What I propose is to have 1. tunnel label - 2. pw label - 3. a label
which combines the pid label + CW + hash label.

The CW is optional anyway.
We could combine the pid and the hash label if we allowed the egress PE to signal multiple pid-label bindings using the proposed procedures, though I have concerns as to how good the LB would be. Clearly the choice of which mapping to use would be a local matter in the ingress PE and the choice of how many mappings to provide would be a local matter for the egress PE. 

- Stewart




-----Original Message-----
From: Luca Martini [mailto:lmartini at cisco.com] Sent: woensdag 18 november 2009 23:18 
To: HENDERICKX Wim
Cc: stbryant at cisco.com; pwe3 at ietf.org
Subject: Re: [PWE3] draft-bryant-pwe3-packet-pw-02.txt

Wim,
We can , of course , always reduce the label stack depth by merging labels, but that is usually not a good idea , as it will increase the number of labels, that that is the most expensive part of MPLS. 
Also note that The FAT PW would not be very useful with a packet PW that

contains IP.
Let assume for a moment that 99.9% of traffic in a packet PW will be IP ... ( most likely it's much higher then that ) 
I do not like the CW much , so let say that we simply encapsulate the IP

packet into a protocol label, a PW label, and a tunnel label.
That's a 3 label stack , with excellent load sharing capabilities. If you pull the PID out into a CW , for example , you will have to add another label , and also make the encapsulation process much more complicated at the PE. 
So , for this application it's better to just have 3 labels.

Luca




HENDERICKX Wim wrote:

I would vote for the latter. It is not much about the label uniqueness
but the depth of the stack since it reduces BW.

If we embed the e-type in the label it becomes more easy as well.

-----Original Message-----
From: Stewart Bryant [mailto:stbryant at cisco.com] Sent: maandag 16 november 2009 18:15 
To: HENDERICKX Wim
Cc: Andrew G. Malis; pwe3 at ietf.org
Subject: Re: [PWE3] draft-bryant-pwe3-packet-pw-02.txt

It depends what you mean by optimize.

Say you have 100 PWs and 2 protocols
If you optimize for minimum number of unique labels by separating the functions, as per the draft, this takes 102 labels 

If you optimize for number of labels in the stack, which is what you

are

requesting, you need 200 unique labels.

- Stewart




HENDERICKX Wim wrote:

This is correct but could we not optimize this special mapping ?

-----Original Message-----
From: Andrew G. Malis [mailto:amalis at gmail.com] Sent: maandag 16 november 2009 15:39 
To: HENDERICKX Wim
Cc: stbryant at cisco.com; pwe3 at ietf.org
Subject: Re: [PWE3] draft-bryant-pwe3-packet-pw-02.txt

Wim,

It's still more efficient on the wire than a simulated Ethernet PW,
because you would have one less label (the protocol identifier), but
would add the Ethernet MAC header instead.

Cheers,
Andy

On Mon, Nov 16, 2009 at 9:03 AM, HENDERICKX Wim
<wim.henderickx at alcatel-lucent.be> wrote:

Ok good, this is what I thought, the client LSP exist in the "client
packet Network layer".

Don't you think this is a huge stack for such a function. You could

have

4-6 LBL(s) + CW (optionally). I understand this proposal re-uses
existing functions defined in MPLS, but there are lot's of

optimizations

possible here -> we could combine FAT LBL, PID LBL, CW potentially.

-----Original Message-----
From: Stewart Bryant [mailto:stbryant at cisco.com]
Sent: maandag 16 november 2009 14:49
To: HENDERICKX Wim
Cc: pwe3 at ietf.org
Subject: Re: [PWE3] draft-bryant-pwe3-packet-pw-02.txt


WH> what I meant is this:

            --------                       --------
   PE1-----|        |                     |        | --- PE3
           |  LSR   |------- PKT PW ------|  LSR   |
           |        |                     |        |
   PE2-----|        |                     |        |---- PE4
            --------                       --------

Assume PE1 has a LSP to PE3 and PE2 has an LSP to PE4, how is this
distinguished on the PKT PW ?



SB> I am not sure what you have drawn there, but it is not what is

shown

in  the draft. The pkt pw run between the PEs and is carried over

the

LSPs. PWs are distinguished from each other at the PE via the PW

label.

WH> What I tried to draw is PE(s) attached to the device (drawn

with

the

box) which is performing the LSR function on a PKT PW to another

device

(drawn with the box) which attaching PE(s).

The question is if I have multiple LSPs from different PE(s)

through

the

PKT PW how are the distinguished. Do I need a separate PID LBL, a
separate PW, etc such that the LSR can make the correct LBL

mappings.

Wim

Think of the PW as a point to point link between a pair of PEs. The

PE

chooses what traffic to sent over a given PW. The PE then pushes an

MPLS

label to identify the destination. The mid-point LSRs have no clue

what

is being carried over the LSP. The egress PE unwraps that packet and
figures out what the payload is.

RFC3985 applies to pkt pw as it does to any other PW.

- Stewart



_______________________________________________
pwe3 mailing list
pwe3 at ietf.org
https://www.ietf.org/mailman/listinfo/pwe3


_______________________________________________
pwe3 mailing list
pwe3 at ietf.org
https://www.ietf.org/mailman/listinfo/pwe3