* Several people still peer via the route servers * Several transits filter their customers by RADB or a private RADB which feeds the IRR.
Care to name names?
route servers: see http://www.rsng.net/ for a list Transits - I did a survey on NANOG a few months ago as to who filtered and how (both peers and customers). About 50% of those who replied used RADB or another similar database (possibly their own, like CA*NET, MCI/CW, Level3 etc.) for filtering either peers or customers. However, I suspect this number is heavilly skewed in favour of vocal NANOG people who like IRRs. Filtering customers was way more prevalent than filtering peers. I said I'd repost the stuff anonymously, but some contributors often post here and thus may chose to answer your question on or off list.
Connectivity failures can and do result when RADB records are not properly updated, which does happen from time to time. They also happened when records WERE properly updated, but the changes made were deemed "too radical" by the software translating the RADB entries into internal databases. Moving a portable prefix from one ASN to another qualified as "too radical" a change, despite it being a semi common occurrence.
And various people had different solutions to this, the most common being a sort of 2 of 3 approach (RADB change, plus sanity algorithm, plus sanity person).
If you do either of the above, chaning a public IRR (once) is easier than changing n private databases. The alternative is no filtering. Hopefully natural selection will take its course on transits who do this on a regular basis.
If common and consistent tools and rules were used to build filters from a SINGLE public database, and if the database site listed contact information and test addresses for each network using the database, I think folks could live with that.
Well if they could agree on a routing policy language, that would be a fine start. -- Alex Bligh GX Networks (formerly Xara Networks)