1) What CAs do we wish to consider as potential issuers for our community? Is it just "Grid CAs" (by that I mean CA we can reasonably except to adhere to best practices as specified by GGF WGs) or do we want to also consider CAs that we have no reasonable expectation of being able to impact their policies or procedures (e.g. commercial CAs) as potential issuers for our community as well?
I think that if we are successful, all this will be used in ways we can't now imagine or, in the future, control. To me, the idea of depending on CA's to issue certificates for DNs that are globally unique is just asking for trouble. Administrative controls to keep the namespaces separate are clearly not good enough. The signing policy file is a technical control but it still seems pretty weak. To me, the thing that is unique is (DN + CA) and the function of the CA is to try it's best to not issue a cert with the same DN to different people. I would be happy if they can do just that and I think it unreasonable to believe that the DN is unique in the universe (or even a small section thereof). The signing policy files basically allow us to say - given this DN, it should have been issued by that CA - and as far as I can see, it's because the CA is't stored in the gridmapfile (and maybe it's not there because the DN was suppoed to be unique - but that was8-10 years ago, and we know better now).
2) Do we believe that during normal operation the CAs indicated in the response to the first question have policy that will result in their issuing globally unique names and will reliably follow that policy?
I think it's not true in "normal operation" and that any moderately talented attacker would be able to generate a condition outside of "normal operations" and get *someone* to issue a certificate with any DN they chose.
3) If a CA is compromised, given currently implementations, this will
(my comments here were in an earlier email).