The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2005-Feb> msg00049



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

[mpls] Last Call on BCP on ECMP

  • From: Erblichs <erblichs@earthlink.net>
  • Date: Sun, 13 Feb 2005 14:19:47 -0800
  • Cc: mpls@ietf.org

Deorge Swallow,

	Sorry, I have a few minor NITs.. The long and
	short possible changes are suggested within..

	All are contained within Current ECMP Practices.


	1) Where we are in a IP environment, it is the
	re-ordering of individual flows where the segments
	in a TCP/IP flows can and do force false fast
	retransmits when multiple duplicate acks are
	recieved. This is due to the fact that one or
	more segments were lost or followed a different
	path that had a longer latency.

	Mis-ordering presumes that out-of-order segments
	can not be corrected at the destination addresses.
	This is not the case in a normal TCP/IP environment.

	Thus a simple change vs the longer one here, is
	to specify the "mis-ordering of flow between" to
	re-ordering of individual flows.

	2) ECMP can only occur when multiple paths exist,
	thus I would suggest "splitting aggregated flows" to
	splitting aggregated flows where multiple paths
	exist.

	3) Due to security purposes, some administrators
	are using intermediate systems that either aggregate
	or modify source addresses. This effectively hides
	the original source addresses and allows the
	intermediate systems to filter incoming non-solicited
	data flows to the original sources.

	Thus, as a FYI. I would possibly suggest ADDING
	something like.

	One should also be aware that some intermediate
	systems are intercepting flows between a source and
	destination and then modifing source
	addresses for security purposes. These modifications
	may alter the effectiveness of of removing ECMP
	routing behaviour. Of course, on the return to the
	modified source address, if the return flow is not
	modified, the flow is then re-directed back to the
	original source.
	

	Mitchell Erblich
	

	

George Swallow wrote:
> 
> This begins a two week last call on
> 
>    Avoiding Equal Cost Multipath Treatment in MPLS Networks
> 
>       <draft-ietf-mpls-ecmp-bcp-00.txt>
> 
>    The last call ends Feb 27 24:00 UTC.
> 
> ...George
> 
> ========================================================================
> George Swallow             Cisco Systems                  (978) 936-1398
>                            1414 Massachusetts Avenue
>                            Boxborough, MA 01719
> 
> _______________________________________________
> mpls mailing list
> mpls@lists.ietf.org
> https://www1.ietf.org/mailman/listinfo/mpls

_______________________________________________
mpls mailing list
mpls@lists.ietf.org
https://www1.ietf.org/mailman/listinfo/mpls