Further into the subject. About the new RIPE-DB software V3.X, is that a requirement when mirroring we have to import ALL the objects in order to satisfy the database referencial integerity ? For example, when we mirror with RIPE, if the route object point to a mntner object (thru mnt-by ) do we have to have that mntner object imported also ? If the answer is YES then does a mirroring agreement mean we have to share all the objects ? Ping Lu Cable & Wireless USA Network Tools and Analysis Group W: +1-703-292-2359 E: plu@cw.net

-----Original Message----- From: Lu, Ping [mailto:plu@cw.net] Sent: Monday, February 12, 2001 10:27 PM To: 'SchmitzJo@aol.com'; andrei@ripe.net; reimp-mig-tf@ripe.net; db-wg@ripe.net; routing-wg@ripe.net Subject: RE: Migration issues: route creation

I think that's me :) Let's talk about the GLOBAL issues and the practical use of RR.

For all RIRs like RIPE, ARIN, APNIC and so on. They are in a regulartory position so they OWN all the objects (AN, PN, MT) and they seldom need to CROSS-REFERENCE with other RIRs. So the self-contained assumption (all references should be resloved locally ) is fine with RIR.

But for the LIRs (especially the ISPs) RR has a more practical use, like auto-generated access-lists and as-path filters. Almost all the major ISPs have their own tools or use the public tool like RAToolSet to pull data from RR then auto-generate the configiuration for their routers.

Now here is the fun part.

As a global player we have to create a list of objects ( AN, RT ) from all the RIRs and with any public peer (registered in one of the RIRs ) we have to create a list of rather private objects ( MT, PN, RO ). But it becomes too difficult to duplicate those private objects with up-to-date information, so the best way is to REFERENCE their LEGAL objects in any one of the RIRs ( of course if the AS itself agree us to do so ). This is where the GLOBALLY UNIQUE nic-hdl comes in and in a more generic sense, how about GLOBALLY UNIQUE maintainer name ?

Does this sound like domain name system ? Sooner or later RR has to face the GLOBAL issue like DNS, IP address and E-MAIL did.

We have already seen some of the global issues showed up in the RR domain ( referral, globally unique nic-hdl ) and there will be more and more name-collision in the future.

So why not adopt a naming system like DNS or E-MAIL ?

Here is my favorite quote from RFC2622 ----------------------------------------------------- <nic-handle> is a uniquely assigned identifier word used by routing, address allocation, and other registries to unambiguously refer to contact information. Person and role classes map NIC handles to actual person names, and contact information. -----------------------------------------------------

Hmmm! So RPSL already specify the need to use information from OTHER REGISTRIES.

I rest my case.

Ping Lu Cable & Wireless USA Network Tools and Analysis Group W: +1-703-292-2359 E: plu@cw.net

...

-----Original Message----- From: SchmitzJo@aol.com [mailto:SchmitzJo@aol.com] Sent: Monday, February 12, 2001 7:32 PM To: andrei@ripe.net; reimp-mig-tf@ripe.net; db-wg@ripe.net; routing-wg@ripe.net Subject: Re: Migration issues: route creation

Dear Andrei,

you wrote that there will be problems creating duplicates of aut-nums and routes in other registries if they do not formally belong there due to the authorization rules introduced with the RPSL transition at RIPE. You suggested two solutions:

...

Possible solutions to this problem are:

1. Send such registration requests to the Database Administration (ripe-dbm) who will perform registration by overriding security (and performing necessary checks before). This puts additional load on ripe-dbm and creates duplicate objects in the RIPE database and "foreign" IRR.

2. Do not allow such registrations. In this case some ISPs will need to change their peering policy and accept routing information from a "foreign" IRR.

Your comments and suggestions are highly appreciated.

I did not yet see an answer to your question. However, I believe that your first suggestion should *not* be followed. I definitely would want to avoid opening "side" paths to enter data into the RIPE database. Using ripe-dbm should be the exception for special requirements. For everything else, the robots should be used.

Regarding duplication: this had originally been introduced to help people along while mirroring was not working. With RIPE having transitioned to RPSL, we will have compatible formats again, and do not need duplication anymore - under the prerequisite that you are properly coordinating with RADB and any other major IRR. I am inclined to say, that avoiding duplication is the purpose of the effort!

Looking at your second proposal, I believe that you are saying the right thing, but phrased it a little different. Peering policy is probably a term which may be confused with routing policy - and nobody needs to change its routing policy by looking at several IRRs now. Actually, I believe that there is not much of an issue: Smaller ISPs have always registered in only one IRR; only bigger providers with international reach have entered their information into several IRRs and duplicated data, but then also using data from all IRRs they registered into. It should not be too much of a pain to use this information to continue with this practice after the RPSL transition.

Those parties who do not agree, speak up loudly now! Thanks Joachim

--- JS395-RIPE -- standard disclaimer ---