On 30 Aug 2025, at 7:49 am, Brian Knight via NANOG <nanog@lists.nanog.org> wrote:
On 2025-08-29 15:15, Tom Beecher via NANOG wrote:
Geoff's data at Potaroo had the FIB clearing 1M back in Febuary if memory serves. So this already happened months ago.
I'm confused by that.
I show 990450 prefixes in my FIB, 990478 in RIB.
Potaroo shows 1022758 prefixes in FIB for today.
Why would there be a discrepancy of over 32k prefixes?
We're blocking exactly 0 prefixes from our three upstreams.
I would understand a hand-waving explanation of "this is the Internet, it's always being updated, and everyone has a different view of it" if the discrepancy were 10^2 or even 10^3. But over 10^5? That's a bit like the three top ASNs for route count just disappeared from the Internet.
every eBGP speaker has a different view of the collection of announced BGP routes. Check out figure 2 of https://www.potaroo.net/ispcol/2025-01/bgp2024.html which plots the 8 hourly RIB count of every peer of Route Views and every peer of RIPE RIS over the 2024/2025 year.s The range of RIB counts varies consistently by 40,000 entries across this set of peers. The IPv6 RIB data has a similar variance of 20,000 entries (out of 200,000). (Figure 15 of the sam article) And, yes, each of these peers has a subtly different set of reachable address prefixes. As of a fre hours ago the "consensus" core spaned some 3,107,033,088 IPv4 /32's in the RIB, but each peer generall cannot see some 50,000 to 150,000 prefixes in its RIB, and sees a further 20,000 to 100,000 /32 prefixes that are not part of a common consensus. So, yes, it really is a case of "this is the Internet, it's always being updated, and everyone has a different view of it", :-) regards, Geoff