Re: GID clarifications

Hi -

> From: "Kanoj Sarcar" <[email protected]>
> To: "Randy Presuhn" <[email protected]>
> Cc: <[email protected]>
> Sent: Wednesday, December 03, 2003 3:04 PM
> Subject: Re: [Ipoverib] GID clarifications


...
> If the provider is managing two unconnected sets of IBA subnets
> (simple example would be a single IBA subnet for A, and a single
> IBA subnet for B, not connected to each other), then two different
> endports might have the same GID; ie, by looking at the GID, you
> could not decide whether it referred to the endport in A or in B.

Now, thinking about a management system that only knows what
it can see through MIBs: how can it know whether A and B are
connected or not?  What happens if they start out unconnected,
and later, due to, for example, a corporate merger, they become
connected?  How do we guarantee that the identifiers for A and B
are unique?  If they are not guaranteed to be unique, does it matter?

> This is true in general, but if everyone decided to always used
> GID index 0, then all GIDs are guaranteed to be unique (I think I
> see what the intent of the statement in the draft was).

This sounds like a safer way to go, but I'm obviously an IB neophyte.

> As Bill pointed out, NAT probably does not look at macaddress/GID
> to make this differentiation. If you believe you have some protocol
> or system that needs to differentiate based on the IPoIB macaddress/GID,
> you should bring it up to the list; it will add weight to the
> argument that everyone must be required to (as opposed to recommended
> to) use GID index 0.

Based on Bill's clarification, I am no longer concerned about NAT.
I am only asking about how the WG intends to manage this stuff
when the identifiers are not globally unique.

Randy