On 3/10/22 07:48, Michael Tremer wrote:
I would strongly recommend this, because parsing the text database and going with the first match is quite likely going to give you inaccurate results.
The reason why we are going with a binary format is because it is organised as a binary tree and therefore can be searched *very* quickly. It also allows us to store multiple networks with the same start address, but a different prefix length which is what you won’t have in a relational database and therefore you will quite likely have inaccurate results.
Yes! I store block announcements w/ autonomous system table foreign keys to support anycast networks and varied prefixes without relation to or dependence on the order in which the set is parsed, albeit at a performance cost as you suggested.
We kept the library as small as possible and we are looking at upstreaming it to all major distributions because that will help integrating it into other software a lot easier. Any help with this would be greatly appreciated.
Very cool, that would be sweet.