In message <[email protected]>
Kireeti Kompella writes:
 
> On May 24, 2012, at 11:40 , Lucy yong wrote:
>  
> > However, the text to be considered as following:
> > 
> >  Load distribution constraint MAY be used during sustained low
> >  traffic periods to 
> >  reduce the number of active component links for the purpose of power
> >  reduction.
>  
> I'd phrase that as "Constrained load balancing MAY ...".  I prefer
> balancing to distribution.

Duh.  Yeah.  Not sure how I managed to mangle the wording so badly and
not notice it so far.  I think this is best:

  Load distribution MAY be used during sustained low traffic periods
  to reduce the number of active component links for the purpose of
  power reduction.

The word "constraint" is dropped.

> In any case, you'd want to add something to the effect that "normal"
> load distribution SHOULD be resumed when traffic increases, with flows
> being affected (again).
>  
> Kireeti.

I see your point except that I don't see this as quite a big a
deviation from "normal" load balancing.

For example, if there are N components, then one could configure the
CL to shut off a component if a condition is met.  One such condition
would be the utilization of the composite would remain under some
configured percentage if traffic remained at the same levels as was
recorded over a prior interval.  For example, if M components remain
(where 1 < M <= N) and traffic was measured in 100 msec intervals for
20 minutes with no interval exceeding a level where M-1 components
would be loaded over 80%, then the number of components is reduced by
one and measurement starts over.  If the composite is ever loaded over
80%, then a component is added back.

That was just an example.  The requirements would just indicate that
some unspecified technique MAY be used to address the requirement.
The framework might put a few more constraints on the technique, but
only if we think that would be needed.  Since coordination on both
sides would be needed to power down components, a protocol extension
would be needed ... eventually, if we could ever get past requirements
and framework.  A bit like LACP only not for Ethernet component links
only, or like Avici Composite Link (abandonned trademark) only not
just for component links using PPP (like POS).

Curtis
_______________________________________________
rtgwg mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/rtgwg

Reply via email to