The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2000-Jul> msg00339



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

Two last calls (draft-ietf-mpls-diff-ext-06.txt)

  • From: Shahram Davari <Shahram_Davari@pmc-sierra.com>
  • Date: Fri, 21 Jul 2000 12:13:11 -0700
  • Cc: Shahram Davari <Shahram_Davari@pmc-sierra.com>, George Swallow <swallow@cisco.com>, mpls@UU.NET

Lou,

What you are saying is that, you want an LSP to span both Diffserv and
non-Diffserv LSRs, but you don't want to use Diffserv edge LSR.

I don't think your assumption that the edge LSR of a Diffserv network being
a non-Diffserv LSR is a valid assumption. Therefore I don't think anybody
can argue about the fact that :

If you have Diffserv nodes in the network, the edge routers must support
Diffserv. 


Regards,
-Shahram


>-----Original Message-----
>From: Lou Berger [mailto:lberger@labn.net]
>Sent: Friday, July 21, 2000 2:59 PM
>To: Shahram Davari
>Cc: 'Lou Berger'; Shahram Davari; George Swallow; mpls@UU.NET
>Subject: RE: Two last calls (draft-ietf-mpls-diff-ext-06.txt)
>
>
>Your example looks fine.
>
>Now consider the case where the ingress/LER doesn't support 
>MPLS DS.  It 
>uses COS1, and sets the EXP bit to 000.  The LSR1 will treat 
>the packets 
>with EXP0, probably DF, while LSR2 will provide COS1 service.
>
>So the introduction of a DS capable transit node can reduce 
>the service 
>seen by an existing  non-DS edge.  This is the problem.
>
>Lou
>
>
>At 11:52 AM 7/21/00 -0700, Shahram Davari wrote:
>>Lou,
>>
>>Let me explain my solution to your problem with an example. 
>Assume we have
>>an LSP that spans LSR1 (Diffserv LSR) and LSR2 (non-Diffserv LSR). Now
>>assume that in the RSVP message we set COS=COS1 and no 
>Diffserv object.
>>Since the LSP is spanning a Diffserv LSR and a non-Diffserv 
>LSR, the LER
>>(Label Edge Router) must be a Diffserv and a non-Diffserv 
>edge LSR. Now
>>assume that the LER sets the EXP field of all packets to EXP=EXP1
>>(indicating a local PHB) in such a way that both EXP1 and 
>COS1 indicate the
>>same local behavior to the LSR1 and LSR2 respectively:
>>
>>
>>                EXP=EXP1   COS=COS1
>>        O----------O----------O------
>>       LER        LSR1       LSR2
>>
>>
>>LSR1 will ignore the COS field and will behave according to 
>pre-configured
>>behavior of EXP1, and LSR2 will ignore the EXP field and 
>behave according to
>>COS1. Since we have designed the EXP1 and COS1 behaviors to 
>be the same,
>>then we will have a consistent behavior throughout the LSP.
>>
>>If the LER sets the EXP to a wrong value (i.e., a value that 
>does not match
>>COS1), then this is a configuration problem, which pretty 
>much exist in all
>>Diffserv networks.
>>
>>Does this make sense?
>>
>>Regards,
>>-Shahram
>>
>> >-----Original Message-----
>> >From: Lou Berger [mailto:lberger@labn.net]
>> >Sent: Friday, July 21, 2000 12:59 PM
>> >To: Shahram Davari
>> >Cc: 'Lou Berger'; George Swallow; mpls@UU.NET
>> >Subject: RE: Two last calls (draft-ietf-mpls-diff-ext-06.txt)
>> >
>> >
>> >Shahram,
>> >         see below.
>> >
>> >At 09:31 AM 7/21/00 -0700, Shahram Davari wrote:
>> >>Hi Lou, George
>> >>
>> >> >PS As mentioned back on 2/8, I believe one trivial solution is
>> >> >to require
>> >> >the use of the DiffServ Object/TLV for all E-LSPs, where use of
>> >> >preconfigured EXP<-->PHB mappings is indicated by either (pick
>> >> >one) no map
>> >> >parameters or a single map parameter set to all zeros.
>> >> >Another solution is
>> >> >to use the one proposed in
>> >> >draft-lefaucheur-diff-te-ext-00.txt, although it
>> >> >does use a new object that could just be merged with the
>> >> >DiffServ Object.
>> >>
>> >>Your solution is ONLY valid if a single LSR could be both a
>> >Diffserv LSR and
>> >>a non-Diffserv LSR.
>> >
>> >It works for both cases, i.e., for the cases where a DS LSR
>> >can support
>> >non-DS LSPs and when it cannot.  In the latter case, a 
>PathErr with an
>> >appropriate code&value would need to be generated.
>> >
>> >>So it could use the presence of Diffserv object to
>> >>decide what behavior it should have (Diffserv behavior or 
>non-Diffserv
>> >>behavior).I don't see any benefit in having a single LSR 
>to be both a
>> >>Diffserv LSR and a non-Diffserv LSR. If the purpose is to 
>support both
>> >>standardized PHBs and some local behavior that are not
>> >standardized, you
>> >>could always use the Diffserv local-PHBs.
>> >
>> >We covered this the last time around.  Here's a relevant excerpt:
>> >
>> >
>> >>Date: Wed, 09 Feb 2000 16:57:57 -0500
>> >>To: Shahram Davari <Shahram_Davari@pmc-sierra.com>
>> >>From: Lou Berger <lberger@labn.net>
>> >>Subject: RE: Announcing draft-ietf-mpls-diff-ext-03.txt
>> >>Cc: "'Lou Berger'" <lberger@labn.net>, "'Jim Boyle'"
>> ><jboyle@Level3.net>,
>> >>         mpls@UU.NET
>> >>Sender: owner-mpls@UU.NET
>> >>[...]
>> >
>> >LB:
>> >
>> >
>> >>> >
>> >>> > One (new) point, your implicit mode hijacks existing
>> >>> > signaling mechanisms
>> >>> > and gives them new meaning.  (Now that I think of it, this is
>> >>> > a large part
>> >>> > of my objection to the current form of the draft.)  For
>> >>> > non-DS nodes, CoS
>> >>> > behavior can be provided by signaling a CoS LSP and not using
>> >>> > the EXP bits
>> >>> > at all.  If that same LSP hits a DS node with the current
>> >>> > draft, it will be
>> >>> > treated as an E-LSP and the matching traffic will receive the
>> >>> > service that
>> >>> > matches bits 000.
>> >
>> >SD:
>> >
>> >>>First of all you have the same situation in a non-MPLS and
>> >Diffserv network.
>> >>>Do you plan to add signaling to them too?
>> >
>> >LB:
>> >
>> >>No because all traffic is best-effort in a DS network.  There
>> >would be
>> >>some work to resolve conflicts if you were trying to support
>> >RSVP on the
>> >>same links/data as DS.
>> >
>> >SD:
>> >
>> >>>Secondly it is a misconfiguration to use a non-DS node in a
>> >Diffserv network
>> >>>and it is very unusual to use a DS-node in a Diffserv
>> >network that supports
>> >>>non-Diffserv behavior too. So why don't we choose the
>> >DS-behavior to be a
>> >>>default behavior in a Diffserv network, and use another
>> >object for signaling
>> >>>those rare non-DS LSPs?
>> >
>> >LB:
>> >
>> >>So you're suggesting a new object/TLV that says "I'm not
>> >diff-serv" rather
>> >>than have the new feature say "I'm diff-serv"?  I don't get
>> >it.  We're
>> >>talking about few bytes of overhead that you've already
>> >defined and are
>> >>willing to incur for all L-LSPs and some E-LSPs.  We're not
>> >talking about
>> >>any substantive change in processing or state.
>> >
>> >SD:
>> >
>> >>>Remember the only situation that you would use a configured
>> >E-LSP is when
>> >>>you are sure that all your nodes are configured and can
>> >support the required
>> >>>PHBs.
>> >
>> >LB:
>> >
>> >>this is not correct.  With the current definition, as soon as you
>> >>configure a node to be a DS node, all established LSPs are
>> >E-LSPs or L-LSPs.
>> >>
>> >>Albeit a corner case for non-green field networks, I think you're
>> >>requiring the configuration of all your nodes at the same
>> >moment of time
>> >>or just not caring about the traffic service levels during
>> >reconfig.  Do
>> >>you view this as acceptable?
>> >>
>> >>Lou
>> >
>> >That's it.
>> >
>> >Lou
>> >
>> >>Regards,
>> >>-Shahram
>> >
>