* I have yet to see a tool which beats human knowledge when it comes to * determining which ranges can be announced as aggregates...
Definately, this is very much an indicator rather than gospel.
Right.
Sure but who is advocating not announcing holes. The problem is, is it possible to ever write a tool to know when an aggregate with holes can be formed or not?
Well, maybe my perception is wrong, and that this is not a problem. I know some people expressed reluctance in doing this earlier. About the tool issue; well, I haven't thought too much about this, but maybe if more than 50% of the address space in a given block (not necessarily on a byte boundary) is routed by the local AS, nobody else would think of announcing the aggregate for the whole corresponding block of address space? As long as all the routed specific routes (possibly from other ASes) are present on the border router of the local AS, it should be safe to announce the aggregate and short-circuit the aggregate (e.g. via a static route) to the null device? Longest-prefix matching takes care of the rest. This is probably as aggressive aggregation as can be done. But again, knowledge of the assignment status may modify this again. I think the Merit tool "aggis" checks for non-routed nearby network numbers to enlarge the suggested aggregation (or was there merely talk of that?). - Havard