The MPLS WG Archive

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

I think you misunderstood me. I said using the first nibble after bottom-most MPLS label to
identify IP header, so that it could be used as part of the hash key, is something new and proprietary (after MPLS was invented) and is not even documented in any informational or standard RFC including RFC2991. After all you could use the MPLS labels only for hash key.

Do you think that all hashing key selection algorithms by all vendors should be supported
by all IETF standards? What if one decides to use CRC to distinguish IP packets? should IETF
standards be backward compatible with that too?

-Shahram

>-----Original Message-----
>From: Curtis Villamizar [mailto:curtis@fictitious.org]
>Sent: Wednesday, March 26, 2003 5:05 PM
>To: Shahram Davari
>Cc: 'Thomas D. Nadeau'; 'George Swallow'; W. Mark Townsley; Andrew G.
>Malis; 'mpls@uu.net'; tnadeau@cisco.com
>Subject: Re: [PWE3] MPLS PID 
>
>
>
>In message 
><4B6D09F3B826D411A67300D0B706EFDE0115C831@nt-exch-yow.pmc-sierra.bc.
>ca>, Shahram Davari writes:
>>  
>> >>1) ECMP could use only label hashing and not hash IP header.
>> >
>> >         This is unrealistic. Lots of vendors hash on both,
>> >one or the other.  Lets not go down the road of trying
>> >to mandate how ECMP works again. We went down that
>> >road in MPLS and the response was clear.
>>  
>> Hashing on IP header is not part of any IETF standard, and I am not
>> sure all fu ture IETF standards should be based on supporting
>> non-standard proprietary implementation s. if one likes to still hash
>> IP header, then he could define a proper protocol mux header and use
>> it all the time.
>
>
>You keep repeating this "non-standard proprietary" diatribe despite
>being corrected.
>
>src/dst based hash has been used for almost 15 years now.  In 1988 you
>might have been more justified in calling it "non-standard
>proprietary" but not in 2003.
>
>It is documented in RFC 2991.
>
>  2991 Multipath Issues in Unicast and Multicast Next-Hop Selection. D.
>       Thaler, C. Hopps. November 2000. (Format: TXT=17796 
>bytes) (Status:
>       INFORMATIONAL)
>
>This makes it far from proprietary and the fact that the document as
>informational does not make it non-standard.  It is not mandated by
>any specific protocol but it is widely implemented and deployed.
>
>src/dst hash based load split is widely deployed in numerous protocol
>and considered important by many ISPs.
>
>So please stop making the "non-standard proprietary" accusation unless
>you want to be known as someone who persistantly ignores the facts.
>
>Curtis
>