Re: please read: Clarification on GUID's universal/loc al bit

The IBTA link WG (who is responsible for this mess in the first place) 
has discussed this issus and is exploring solutions.

The current thinking is not to use the u bit to indicate global scope. 
Instead, any locally administered IB addresses (i.e. IB 64-bit Global 
identifiers) would use a reserved vendor identifier (i.e. a reserved 
OUI).  Any IB 64-bit Global ID that does not use this vendor tag will 
have global scope and be globally unique.

In this approach, the IBTA would apply to the IEEE Registration 
Authority for a OUI to be used for this purpose. The IBTA would also 
clarify the IBA spec and add the requirement the "u" bit is ignored in 
the EUI-64 (i.e. a vendor cannot have two EUI-64s differing only in the 
value of the "u" bit)

Diego Crupnicoff wrote:
> Hi Margaret,
> 
>  > (2) Accept that we cannot tell, by looking at an IB GUID, whether it
>  > is of global or local scope.  In this case, the conservative choice
>  > would be to treat all IB GUIDs as if they are of local scope, always
>  > clearing the "u" bit in the IPv6 IID, regardless of how the bit is
>  > set in the IB GUID.
> 
> I think it may actually be safe to assume that the IB GUID in question 
> is __always__ globally unique (regardless of the value in its "u" bit). 
> The IB spec says:
> 
> "The manufacturer assigns an EUI-64 __with_global_scope_set__"
> 
> The IPv6 IID can therefore be based on this manufacturer assigned GUID 
> which is __required__ by the IB spec to be one with global scope. The 
> mechanism for building the IPv6 IID could then be as simple as taking 
> the IB mfg assigned GUID and set (if it was not already so) its "u" bit 
> (i.e. ignore the interpretation that the vendor may have given to the IB 
> spec).
> 
> To close the loop, the IBTA would have to explicitly forbid the use of 
> the same 63 bit combination with two different values for the u bit as 
> two valid globally unique GUIDs. I can't speak for the entire IBTA but 
> it looks to me that this would be acceptable. Furthermore, it is my 
> understanding that the definition of locally unique IB GUIDs would then 
> be irrelevant to the ipoib spec.
> 
> Does it make sense?
> 
> Thanks,
> 
> Diego
>