Dear grow members, please find the minutes of the Peering API side meeting
below:

Attendees: Aussie Broadband, Amazon, Workonline, Huawei, APNIC, Telstra,
Meta, Cloudflare (12 attendees)

1. Brief walkthrough of PeeringAPI.
2. Open questions: PNIs: who issues LOAs, who orders XCs, exchanging
preferences (redundant links, who unfilters first, ...). PeeringDB is not
good enough. RPKI signed checklists could be worth it. Provisioning Finite
State Machine.

3. Work with Peering Manager to implement the API to advance adoption.
4. Aussie agrees this is a good idea. Great for an eyeball network. Reduces
complexity, single pane of glass.
5. 2 Types of business logic: peering logic and business logic that goes in
the provisioning. The second one is what needs modelling.
6. IXP Manager integration?
7. Are there any additional security considerations?
8. Ben Maddison explains RSC. Signatures over arbitrary blobs of data.
Rough workflow: each party issues a challenge. Sign the challenge. Provide
signed blob back over the API (inband).
PeeringDB is good enough for now, but baking it into the protocol as a
first-class citizen might be a mistake. Make RSC a first-class citizen, and
machine-to-machine OATH as a fallback. Similar to the key negotiation
within SSH: Offer list of IdPs, receive list of IdPs. Ordering is not
important, as long as you pick an IdP that is trusted. No need to use the
same IdP in both directions.
9. Don't roundtable the FSM. Might need a workflow negotiation system.
Might be good to communicate operational state. Get a feedback loop going.
2 classes of state transition: 1 that requires coordination, 1 that
doesn't. To what extent do we expose "hygiene" of turnup testing. Protocol
coordination will need to happen for ordering-of-actions. Might need a
deadlock state.
What are the preferences we should care about? Provide a human address for
"escalation" in case of deadlock.
Tie-break decision algorithms?
Make the errors more expressive, but structured text. ENUM for the most
frequent ones, but allow extendability. Look to YANG?
Allow for resumption through the FSM once deadlocks have been resolved. If
data-leaf is not provided, block state transition, and progress once that's
there. Coordinated data-collection exercise.
Route-server sessions?

Next steps: We'll work on adding a section on the RSC challenge-response
proposed authorization and authentication workflow. Work will also start on
a minimal provisioning FSM. We want to thank all the contributors of
today's meeting, and look forward to working with the Working Group in
advancing this draft.
-- 
Tom Strickx
Principal Network Engineer
AS13335 - Cloudflare
_______________________________________________
GROW mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/grow

Reply via email to