Re: Limited Domains:

["Ahmed Abdelsalam (ahabdels)" <[e-mail]>]

Inline [AA]

-----Original Message-----
From: Tom Herbert <[e-mail]>
Date: Tuesday, 13 April 2021 at 00:21
To: Brian E Carpenter <[e-mail]>
Cc: Fred Baker <[e-mail]>, "[e-mail]" <[e-mail]>, "[e-mail]" <[e-mail]>, ahabdels <[e-mail]>
Subject: Re: Limited Domains:

    On Mon, Apr 12, 2021 at 3:10 PM Brian E Carpenter
    <[e-mail]> wrote:
    >
    > On 13-Apr-21 08:52, Fred Baker wrote:
    > >
    > >
    > >> On Apr 12, 2021, at 1:51 PM, Brian E Carpenter <[e-mail]> wrote:
    > >>
    > >> In other words, a redefined flow label could be used pairwise between hosts that have determined that they are both members of a particular limited domain.
    > >
    > > Which would eliminate the use of the flow label by systems that are not part of that domain.
    >
    > No, it would create two sets of nodes: nodes using RFC6437 between themselves, and nodes using some other interpretation of the flow label between themselves. Ships in the night, really just like running IPX and CLNP on the same network. We used to do that a lot.

    Brian,

    The problem is that there's no code point that  discriminates between
    the different interpretations of the flow label. So in a limited
    domain the deployment of the new implementation must be perfect, lest
    there is at least one router that is inadvertently run with the legacy
    interpretation, then that router may process the flow label and ECMP
    per the standard leading to situations where packets of some flow are
    unexpectedly going over different paths. Since there's nothing in the
    packet that states what's happening, such problems can be really
    difficult to debug.

[AA] I don't think we need a new codepoint.
The operator explicitly configures the routers under his administrative domain to support Structured Flow Label. Note that all the nodes involved in the IPv6 packet lifecycle (source node, transit nodes, destination nodes) ALL are under the same administrative domain of the operator. If there is a router that does not support the Structured Flow Label, then the 4FLC bits are NOT set/inspected by the source/transit node.
Keep in mind that packet reordering is part of the SLA in between the operator and their customers. The operator will not enable something that breaks their SLAs on purpose.

    Tom

    >
    > I'm not advocating this, just pointing out that it's possible.
    >
    >    Brian
    >
    >
    > --------------------------------------------------------------------
    > IETF IPv6 working group mailing list
    > [e-mail]
    > Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
    > --------------------------------------------------------------------