> It says a host
> MAY set the flow label, for instance, rather than what it MUST do. I
> think it would be better for the document to say a HOST just sets it
> to 0. Or, if that is not right, what it SHOULD/MUST do. We want
> predictable behavior from the originating hosts.

But we can't do that realistically. A host may choose to set the flow
label on no traffic at all, on particular types of traffic that it cares
about, or on all traffic (unlikely). This will be a matter of local policy
configuration, and/or the level of QOS support available in that particular 
version of that particular operating system. 

What this document attempts to say is what should happen in cases where hosts
*choose* to activate the flow label. So the whole document is inevitably
a MAY. Perhaps it would be better if that was explicit - an applicability
statement written as a MAY, but SHOULDs in the body of the text to define
the desired behaviour.

Thanks for reading it anyway!

   Brian


Thomas Narten wrote:
> 
> Hmm. There was much discussion in SLC on the Flow Label.  Quite a lot
> of mail in the weeks immediately after SLC. Then quiet. Then a new ID
> appears. More silence. Something like 45 minutes of agenda time has
> been allocated next week for the topic. Has no one had time to read
> the draft yet? :-)
> 
> Here are some high-level comments for folks to think about.
> 
> The scope of the document could be narrowed. I think its focus might
> better be just what a host and router MUST do in order to be
> compliant. There is a fair amount of additional text that talks about
> MAYs with regards to possible usages.
> 
> It could say more about what a host is supposed to do. It says a host
> MAY set the flow label, for instance, rather than what it MUST do. I
> think it would be better for the document to say a HOST just sets it
> to 0. Or, if that is not right, what it SHOULD/MUST do. We want
> predictable behavior from the originating hosts.
> 
> In fact, the document might just have too many MAYs. It seems like it
> is trying to not make it illegal for someone to experiment with the
> flow label. I'm OK with that, but that doesn't need to be put in the
> spec as MAYs. It should be clear what RFC-compliant implementations
> are supposed to do. Not sure it needs to say anything more. People are
> always free to experiment.
> 
> There is perhaps too much wording about what signaling (and "methods")
> should do with the flow label. IMO, much (all?) of this is future
> work, and this document shouldn't constrain it in anyway. For example,
> there may well be too much wording about how one might use the flow
> label. For example, there is wording about allowing the receiver to
> tell the sender what flow label should be used. This seems premature.
> 
> There is even wording that suggests classifiers need to handle
> wildcarding of the source and dest fields in the tuple. Not sure how
> much of that is appropriate.
> 
> In summary, I think it would be good for this document to make it
> clear exactly what compliant hosts and routers MUST do. Trying to say
> anything more than that may tending towards speculation that is
> premature.
> 
> Thomas
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------

Reply via email to