AS208147 is live: watching global convergence happen
Follow-up to Starting to set up AS208147.
The BGP session with the first transit provider is up, both prefixes are announcing, and β as of today β the whole thing has fully converged globally. Here’s what that actually looked like from the inside.
The two-stage build
Every BGP setup really has two separate steps, and it’s worth treating them as genuinely separate rather than one action:
- Get the session established. Peer with the transit provider, confirm the connection is alive, and β deliberately β announce and accept nothing yet. Filters set to reject everything in both directions. This proves the plumbing works before touching anything that actually affects routing.
- Open it up. Add the static routes, tell BGP what to actually advertise, swap the filters from reject-everything to the real accept/deny logic. Only once step 1 is verified stable.
Doing it in that order meant every problem showed up in isolation, one layer at a time, rather than all at once.
Watching it propagate
This was the part I hadn’t really appreciated before doing it live: BGP propagation isn’t instant, and it isn’t uniform. Some networks pick up a brand-new announcement within minutes β anywhere with a short, direct peering relationship converges fast. Others depend on scheduled IRR-derived filter refreshes that might only run once a day, so a first-ever announcement can take a genuinely long time to reach full global visibility, even though the underlying protocol itself is fast.
Watching this in real time (RIPE Atlas measurements, bgp.tools, BGPlay, and a lot of hop-by-hop traceroutes) was a good reminder that “the internet has converged” isn’t a single moment β it’s dozens of independent networks each catching up on their own schedule. I could ping most of Europe within an hour. It took a lot longer for anywhere else.
From first announcement to full global reachability: roughly 11.5 hours. The last holdout was, painfully, one of Google’s own resolvers (2001:4860:4860::8888, the IPv6 counterpart to the famous 8.8.8.8) β everything else, including a genuinely wide geographic spread of RIPE Atlas probes across every populated continent, had already come online well before that.
Confirmed, from the outside
The most satisfying moment was checking bgp.tools and seeing both prefixes listed cleanly:
Originated by AS208147 β Evie Snuggle
Both 44.32.187.0/24 and 2001:67c:1970::/48, visible, correctly attributed, and independently confirmed reachable from real hosts on real networks around the world.
What’s next
Same plan as before: the second site, second transit provider, and the two talking to each other. Now that there’s a known-good process for standing up one session β two-stage build, verify before opening, expect a first announcement to take the better part of a day to fully settle β the second one should go a lot faster.
Starting to set up AS208147
Snuggle Systems now has its own slice of the internet’s routing table: AS208147 is officially registered, and I’m starting the process of actually putting it to work.
What’s actually happening here
An ASN doesn’t do much, it only becomes useful once you start using it to announce via BGP towards upstreams willing to carry the announcements. So the plan, roughly:
- Two independent sites: each with its own path to the internet, so a failure in one location doesn’t take the network down.
- Two transit providers: at least at the beginning, one per site with the two sites talking to each other directly.
- Two IP address prefixes: one /24 for IPv4 and one /48 for IPv6
What has been done
The boring stuff: paperwork, applying for an ASN, getting PI (provider-independent) IP address space, talking to transit providers, getting RIPE database objects in order, setting up RPKI ROAs, setting up IRR objects, making a PeeringDB entry so that AS208147 is discoverable and contactable by anyone who needs to reach it, and finally, making this page.
What’s next
Next up: the actual BGP sessions - getting the two sites talking to their respective transit providers, then to each other. I’ll hopefully do a follow-up once it is working.
If you’re building something similar and want to compare notes, feel free to reach out to shiny new AS208147!