[email protected]
[Top] [All Lists]

RE: FW: composite link - candidate for respin, maybe

Subject: RE: FW: composite link - candidate for respin, maybe
From: Yong Lucy
Date: Wed, 14 Apr 2010 10:04:44 -0500
Hi Curtis,

Very glad to see our discussions driving toward the same understanding.

See inline.

Thanks,
Lucy

> -----Original Message-----
> From: [email protected] [mailto:[email protected]]
> Sent: Wednesday, April 14, 2010 1:24 AM
> To: Yong Lucy
> Cc: [email protected]; [email protected]
> Subject: Re: FW: composite link - candidate for respin, maybe
> 
> 
> In message <[email protected]>
> Yong Lucy writes:
> >
> > Hi Curtis,
> >
> > Snipped:
> 
> 
> Thanks for trimming.
> 
> 
> > > The conclusion was that at a midpoint LSR the LSP defines the top link
> > > but the entire label stack can be used for load balance.
> > >
> > [LY] IMO: We may need to clarify what "load balance" means first. Do you
> > mean load balance over all component links? If component links have
> > different metrics, load balancing over all component links may not meet
> > operation requirements. In addition, load balance is typically referred
> to
> > as distributing traffic without BW reservation. In the context of
> previous
> > requirement draft, it requires that traffic placement function have BW
> > reservation capability. Hope other co-author can help clarify on this.
> 
> 
> Load balance means to put traffic on component links of the composite
> link such that the component links are evenly balanced to some
> approximation of even.
[LY] Agree.
> 
> It is possible to mix hard reservations and soft reservations and do
> so on a LAG or CL or multipath or whatever you'd like to call it.
[LY] load balance over LAG and multipath (ECMP) is common practice today. If
composite link contains component links that have different characteristics,
i.e. traffic over different component links gets different performance, load
balance may not meet carrier service requirement. If all component links in
a composite link have the same performance, then load balance can apply over
it. 
> 
> If you want the benefit of statistical multiplexing you have to live
> with not having everything done in hard reservations.  Most of this,
> but maybe not all can be done with existing protocol bits.  With smart
> use of QoS you can have high priority services with enforceable hard
> reservations sharing bandwidth with other traffic whose reservations
> are soft of "advisory" and statisitcally very likely to be met but not
> guarenteed.  The first has an SLA.  The second doesn't and pays much
> less but in practice often gets the same service quality.
[LY] This is a solution for LAG and multipath. Like you said, existing
protocol may not support coherent function well.
> 
> 
> > Snipped
> > >
> > >
> > > I don't think we agreed not to include IP if it took very little or no
> > > effort to include IP relative to LDP.
> >
> > [LY] In my opinion, the previous requirement draft addresses both kinds
> of
> > IP traffic, but not addresses other IP traffic.
> 
> 
> Please explain.  Both kinds but not other kind?
> 
> 
> > Snipped,
> > >
> > > You have that wrong.  In the composite link requirements a LSP can
> > > have a bandwidth that is greater than any one of the component links.
> > >
> >
> > [LY] This is new to the previous requirement draft. It is OK with me if
> the
> > new version has this requirement. I see the merit. Thanks.
> 
> 
> Yes it is a requirement that we agreed to in Anahiem.
> 
> 
> > snipped
> > > > [LY] Yes. However, what is the flow for composite link in latter
> case,
> > > > TE LSP or LDP?
> > >
> > >
> > > It doesn't matter how an LSP is set up.  The same techniques that just
> > > look at the label stack (or optionally past the label stack if the
> > > payload is IP) work regardless of how the LSPs were set up.
> > [LY] This also have some new aspect to the previous requirement draft.
> >
> > Lucy
> 
> 
> Curtis

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

<Prev in Thread] Current Thread [Next in Thread>