[138541] in North American Network Operators' Group
Re: L3DSR server side bits open sourced
daemon@ATHENA.MIT.EDU (Igor Gashinsky)
Wed Mar 9 13:33:42 2011
Date: Wed, 9 Mar 2011 13:30:37 -0500 (EST)
From: Igor Gashinsky <igor@gashinsky.net>
To: Shane Amante <shane@castlepoint.net>
In-Reply-To: <2A59479C-DD1E-4A1D-9719-44CA77E16DAC@castlepoint.net>
Cc: nanog@nanog.org
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
On Wed, 9 Mar 2011, Shane Amante wrote:
::
:: On Mar 9, 2011, at 00:35 MST, Igor Gashinsky wrote:
:: > On Wed, 9 Mar 2011, Randy Bush wrote:
:: >
:: > :: a real use for the diffserv bits! why not flowlabel in 6? it's been
:: > :: looking for a use for a decade.
:: >
:: > Honestly, we figured flowlabel might actually find a use before all the
:: > values of diffserv will :) In all seriousness, we are starting to set the
:: > spec for v6 l3dsr now, so, if you care, and believe that flowlabel would
:: > be a better field to "hijack" (or have a suggestion for another, better
:: > way then same DSCP methodology that we used for ipv4), we welcome input..
::
:: :-/ Please don't abuse the flow-label this way, otherwise your proposal could get added to the "graveyard of IPv6 flow-label proposals" draft:
:: http://tools.ietf.org/html/draft-hu-flow-label-cases-03#section-3
::
:: There has been *a lot* of discussion in the 6man WG recently to (finally) define the flow-label to be: a) be stateless; and, b) potentially be useful as an input-key, when used in conjunction with {src_ip, dst_ip}, for fine-grained load-balancing over LAG & ECMP paths, (instead of the traditional IPv6 header 5-tuple). One example where this might be useful is within Layer-2 switches, at IXP's or other parts of the network, where you'd really like them to only have to look at the 3-tuple: {src_ip, dst_ip + flow-label} as input-keys for LAG load-balancing, since they are at a fixed location in the IPv6 header. The other, longer-term win of this approach is that hosts can be free to define, or re-define, new IPv6 Extension Headers and you won't have to worry about Core routers/switches needing to dig into those Ext. headers (or, past them) to find useful input-keys for load-balancing over LAG & ECMP paths.
::
Yeap, this is why I said flow-label might actually find a good use soon
enough :)
-igor
