[openib-general] IB Address Translation service

Hal Rosenstock halr at voltaire.com
Tue Apr 5 14:02:14 PDT 2005


Reviving an old thread...

On Sat, 2005-03-05 at 10:22, David M. Brean wrote: 
> There is an I-D for DHCP on IB.  IPoIB defines a "broadcast" address and 
> DHCP (and ARP) on IB use it.  Could make RARP work using this mechanism, 
> but as someone else pointed out, the IB hardware address contains a 
> QPN.  The I-D for IPoIB says something like:
> 
>     The link-layer address for IPoIB includes the QPN which might not be
>     constant across reboots or even across network interface resets.
>     Cached QPN entries, such as in static ARP entries or in RARP servers
>     will only work if the implementation(s) using these options ensure
>     that the QPN associated with an interface is invariant across
>     reboots/network resets.
> 
> So, there are requirements on the IPoIB implementation to make RARP 
> work.  Folks in the IPoIB work group decided not to go much further than 
> these statements for RARP support since most folks felt that DHCP is (de 
> facto) replacement.

There are 3 cases I can envision:

1. A single IPoIB interface per HCA port. In this case, the RARP server
can just match on the hardware address (port GID) without the QPN.

2. In the case of VLANs, I think we are likely OK as well. In that case,
there is a separate IP subnet (per PKey) so the port GID is unique per
IP subnet (the port GID is unique on that partition (IP subnet)). I
think there is a different QPN per VLAN.

So I don't think that the above 2 cases require an invariant QPN.

3. The third case is multihomed interfaces on the same IPoIB subnet. I
don't think this is currently supported by IPoIB (but may someday). That
would either not be supported by RARP or some way to have invariant QPNs
would be needed. I'm not sure how important this case is.

Is the above correct ? Are there other cases ? 

-- Hal



More information about the general mailing list