Re: [manet] DLEP LID lengths

[Stan Ratliff <[e-mail]>]

On Mon, Jun 19, 2017 at 11:53 AM, Rick Taylor <
[e-mail]> wrote:

> Comments inline...
>
> On Mon, 2017-06-19 at 10:34 -0400, Stan Ratliff wrote:
> >
> >
> > On Mon, Jun 19, 2017 at 10:20 AM, Rick Taylor <rick@tropicalstormsoft
> > ware.com> wrote:
> > > All,
> > >
> > > I've been mulling this over.
> > >
> > > How about the LID length MUST be the same as the MAC, so it's up to
> > > the
> > > advertiser of the LID to bit pack?
> > Does that mean that if a modem is using EUI-64 MACs, then the LIDs
> > are 64 bits as well? I'm OK with that, I just think we need to state
> > it.
>
> Yes, that's the idea.  A draft-01 will be forthcoming...
>
>
> > Also, I think there should be a sentence about LIDs not being
> > duplicated (and, in fact, that a LID MUST NOT duplicate an existing
> > MAC, or vice-versa). I'd think that a duplicate LID on Destination Up
> > would be rejected, and the Destination would not be seen as Up
> > (Destination-level error, NOT a peer session level error).
>
> I'm not sure about this.  I can see why it seems helpful:
> an implementation can keep one data structure for all destinations,
> physical/logical/LID.  But, I can't see a simple way for a dual-mode
> modem implementation to ensure no clashes.
>
> Part of the reasons behind LIDs was to produce a separate namespace.
> Forcing and implementation to maintain 2 datastructures (of the same
> type) for the two namespaces seems little extra effort (again, ignoring
> ADA implementations).
>

Yes, implementations could maintain 2 information bases. But especially
since the LID length will be the same as MAC length, it's an easy extension
to put both in the same tree.
As for how to keep them from clashing - both modems and routers have to
maintain a database (or set of in-memory structures) to be able to report
stats and keep track of what's going on. Since LIDs are "made up", while
MACs are "architected", it would seem pretty easy to:
1. Make a LID up
2. Check to see if that's already in the tree, and if so,
3. Make another one up...

I could even see a pretty straightforward implementation where LIDs are
assigned based on a monotonically increasing counter. I've never see a MAC
address of 00:00:00:00:00:01, 00:00:00:00:00:02, etc... however the
implementation decides to pack the LID into a 48-bit (or 64-bit) quantity.

Regards,
Stan



>
> >
> > >
> > > This seems to work with existing use cases, and avoids the need for
> > > LID
> > > length negotiation.  The only down-side I can see is that 2^48 LIDs
> > > is
> > > a common maximum, but that seems plenty currently.
> > Yeah, I'm thinking that 1 modem supporting 2^48 links will cause
> > insanity well before DLEP has a problem... ;-)
>
> As was pointed out to me: 640Kbytes has traditionally been plenty, 2^48
> looks like carelessness! :-)
>
> Cheers,
>
> Rick
>