The MPLS WG Archive

[Shahram Davari <Shahram_Davari@pmc-sierra.com>]

Hi Sasha,

The Diffserv over MPLS draft is not introducing any changes to Diffserv. Therefore all the Diffserv standards must be followed including not reordering the packets belonging to a microflow.

It is true that in theory it is possible to reorder packets that belong to a microflow if PHBs are not mapped properly to E-LSPs, but that is a misconfiguration. I think everybody knows that they should not misconfigure their network.

I don't think we need to tell people how to avoid misordering, all we have to do is to require that no misordering to happen.

Just as an example assume that an AF1x aggregate consists of only 2 microflows: A, and B. If A never exceeds its committed policed rate (therefore all packet belonging to A are AF11), and if B's committed rate is zero, in other words it always exceed its committed rate (therefore all its packets are AF12/AF13), then no reordering of microflow will happen if we send the AF11 aggregate through an E-LSP and the AF12/AF13 through another E-LSP.

Not that I recommend such a configuration, but the above example shows a scenario that is actually compliant with Diffserv.

Yours,
-Shahram

   



> -----Original Message-----
> From: Sasha Vainshtein [mailto:Sasha@AXERRA.com]
> Sent: Thursday, April 05, 2001 12:16 PM
> To: 'N S S Kishore Kankipati'; flefauch@cisco.com; liwwu@cisco.com
> Cc: mpls@UU.NET; 'diffserv@ietf.org'
> Subject: RE: Diff-Serv over MPLS
> 
> 
> 
> Hi,
> 	The reordering issue has been addressed in
> <draft-ietf-diffserv-new-terms-04.txt>,
> 	"New Terminology for Diffserv" with an explicit 
> reference to work 
> 	on MPLS Support of Differentiated Services.
> 	 Section 5 of this I-D introduces the notion of an Ordering
> Aggregate (OA) as 
> 	<quote> A set of Behavior Aggregates that share an ordering
> constraint. 
> 	 The set of PHBs that are applied to this set of 
> Behavior Aggregates
> 
> 	constitutes a PHB scheduling class. <end of quote>
>      	
> 	I suggest that <draft-ietf-mpls-diff-ext-08.txt> should 
> be updated
> to 
> 	allow only grouping of multiple OAs (and not BAs as it 
> does) into
> 	a single E-LSP. The current I-D version explicitly states
> 	that L-LSP is established for a given (OA, FEC) class but is  
> 	ambiguous when it comes to E-LSPs. 
> 
> 	Nabil, in his message in this thread, has noticed that RFC 2597
> 	defines AF<x><y> (y=1,2,3) as an OA for any fixed value 
> of x (1-4). 
> 	The suggested modification would prohibit 
> configurations like one
> described
> 	in the original message. 
> ______________________________________________
> With best regards,
> Sasha Vainshtein
> mailto: sasha@axerra.com
> phone: +972-3-7659993 (office)
> fax:      +972-3-6487779 (office)
> 
> 
> >>-----Original Message-----
> >>From: N S S Kishore Kankipati [mailto:kishore_iitb@yahoo.com]
> >>Sent: Thursday, April 05, 2001 3:39 PM
> >>To: flefauch@cisco.com; liwwu@cisco.com
> >>Cc: mpls@UU.NET
> >>Subject: Diff-Serv over MPLS
> >>
> >>
> >>hi all	
> >>   Could any of you please explain about the following
> >>sentence in the draft-ietf-mpls-diff-ext-0.8.txt.
> >>
> >>section 1.2-- Exp-Inferred-PSC LSPS (E-LSP)
> >>
> >>"A single LSP can be used to support up to eight BAs
> >>of
> >>a given FEC, regardless of how many OAs these BAs
> >>span."
> >>
> >>A small doubt
> >>  Suppose PHBs related to a PSC have been assigned to
> >>different E-LSPs (say AF11 to one LSP and AF12 to
> >>another LSP)and first few (say 1-10)Packets belonging
> >>to a microflow have been assigned to AF11 and later
> >>packets (say 10-15) to AF12. 
> >>Is it not possible for the packets in AF12 LSP to get
> >>transferred before the packets in AF11 LSP? If that is
> >>the case the ordering constraint is violated i.e
> >>packets are reordered. (I think this may happen in 
> >>the following case- if the LSPs are on different 
> >>interfaces of a router and  there may be some packets
> >>which are required to be served in the AF11 while the
> >>packets in AF12 may get immediately served as there
> >>are
> >>no packets before these which require service).
> >> 
> >>thanks
> >>kk
> >>
> >>__________________________________________________
> >>Do You Yahoo!?
> >>Get email at your own domain with Yahoo! Mail. 
> >>http://personal.mail.yahoo.com/
> >>
>