Hello all (special call out to network operators),

Suppose a network operator (AS) makes an error in the creation of their ASPA - 
they did not include the AS of one of their providers.  Perhaps this would be 
rare.

If you are a provider and this error was by your customer's customer or earlier 
in the path, then there is no issue. The ASPA verification method at your AS 
will evaluate the route as Invalid. That is fine.

Consider the case when this error was made by your *direct customer*, and you 
can see it is a mistake. In your border router, you have configured them as a 
customer. You may have also confirmed it by exchanging BGP Role with the 
customer [RFC 9234]. The existing ASPA algorithm determines the 
customer-to-provider relationship by your local configuration and does not pay 
attention to the immediate customer's ASPA. This is because the locally 
configured BGP role has precedence and the creation of ASPA is informed or 
driven by that. So, by using the existing ASPA algorithm, your AS will not flag 
the customer routes as Invalid due to the above-mentioned error in their ASPA.

Is the above fine with you (as the provider)?  The advantage is as follows:  
Say, the customer is in good standing with an established BGP session. They may 
be stub/single homed AS (a significant % of ASes in the Internet). You have 
provided connectivity to them over time and then one day they create their ASPA 
with an error (forgetting to include you). You do not immediately cut them off 
from the Internet (based on the current ASPA algorithm).  Instead, you maintain 
reachability for them and fix their ASPA error out-of-band. You detect, notify, 
and help fix the error out-of-band (OOB) by a suitable operational procedure.  
Only those neighbors of yours who are your provider or lateral peer and doing 
ASPA will detect the route as Invalid but all other neighbors (including all 
you customers and their customers) will not be affected by the error and will 
continue have reachability to your customer and their customers. Of course, 
when the ASPA error is fixed (hopefully quickly), everyone would be fine.

The idea is not to ignore the error in the immediate customer's ASPA. OOB 
procedures would be in place to help the customer fix it quickly. Also, when 
ASPA deployment starts to gain momentum, the network operators would likely 
have automated tools to check their ASPA while they are registering and 
discover any errors. This could make the probability of ASPA errors negligible 
or zero.

The alternative is for the designers (authors/WG) to bend over backwards and 
make some substantial changes in the ASPA verification algorithm to strictly go 
by the ASPA and give it precedence over the local configuration set by you (the 
provider).  The erring customer's routes will be rejected by you. Also, the 
resulting more complex procedure in this case does not give you a choice to 
override the ASPA verification outcome and maintain connectivity for the 
well-established but erring customer.

OTOH, if the existing ASPA algorithm is kept unchanged, then the provider still 
has the option to detect the ASPA error OOB and terminate the BGP session if 
they wish to use that approach to force the customer to fix the ASPA error.

Should IETF specification go the extra mile to have more complex in-band 
procedures to compensate for operational (or implementation) errors or it is 
better to have OOB operational procedures to catch and deal with such errors?

Sriram





_______________________________________________
GROW mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to