Hello Sriram and all,

Thank you for raising this important operational scenario. As a protocol
designer (not a network operator), I believe intrinsic detection of ASPA
registration errors should be within the verification framework itself –
especially for direct customers. My rationale is threefold:

1. Early error detection is critical – *Out-of-band solutions increase
complexity*

The core issue is: *How does a provider proactively discover a customer’s
ASPA registration error? *

Under today’s algorithm, providers have no *in-band mechanism* to detect
missing ASPA entries from direct customers. This creates "silent failures"
where reachability degradation (from peers/providers) might go unnoticed
until complaints arise.

By revising the verification algorithm (e.g., prepending the local AS and
perform egress verification), providers gain immediate visibility into
misconfigurations. An automated *out-of-band solution for detecting
registration errors obviously increases the complexity of deployment*.


2. *Not the ideal connectivity while masking the errors*

When a customer’s ASPA omits their provider (e.g., local AS), today’s
algorithm creates a validation gap: Local AS sees no error locally (BGP
role takes precedence), But upstream ASes flag the route as Invalid due to
ASPA checks.

This gap causes asymmetric reachability:
Local AS ───[BGP: Valid]───>>> Customer
 │
 │ ASPA check by others
▼
Upstream ASes ───[ASPA: Invalid]───✘ Customer (Silent drop)

Result: The customer is "connected" to local AS but disconnected from
peers/providers – * not the ideal connectivity that operators expect, while
also masking the issue*.


3. ASPA's Internal Safeguards *Preserve Connectivity While Enforcing
Correctness *

If partial connectivity is critical for network operators, strict ASPA
checks need not result in immediate disconnection. We can:

(1) Treat *"Invalid" at egress* as a soft alarm – Flag errors but allow
traffic flow. Network operators can choose for themselves whether to
strictly filter "Invalid" routes or just treat them as alerts; this is a
simple deployment and configuration approach.

(2) Set up a short-term whitelist until the registration issue is resolved.

These thoughts reflect my perspective as a protocol designer. Thank you,
Sriram, for raising this issue. I'm particularly keen to hear feedback from
network operators.

Best regards,
Jia
Zhongguancun Laboratory
_______________________________________________
GROW mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to