Hi Gyan,
Thank you for your comments. Welcome to join the work on this draft. I think we
can further discuss and clarify the scenarios of the draft.
Best Regards
Yisong
发件人: Gyan Mishra
时间: 2023/10/08(星期日)06:22
收件人: SPRING WG;draft-liu-spring-sr-policy-flexible-path-selection;
主题:
https://datatracker.ietf.org/doc/html/draft-liu-spring-sr-policy-flexible-path-selection
review
Dear Authors
I reviewed the draft presented at IETF 117 and had some questions relating to
the draft.
SRv6 with it’s simplicity does not require SR policy VPN color overlay to SR
policy intent mapping to instantiate the data plane forwarding as does SR-MPLS
so generally this does not come into play unless you are required to
instantiate data plane using a static sid list.
If ODN (On Demand next hop) or automated steering is utilized in the SR policy
then as well for both SR-MPLS and SRv6 do not require to use static SID
list static candidate paths.
If you have many PEs for simplicity with stateful PCE delegation it is simpler
for operators to use ODN and automated steering techniques.
The main focus for this draft is for use case when using a static SID list for
the candidate path and have multiple static SID lists with the same priority
how the best path candidate is selected which today it’s static via priority.
The draft mentions that a candidate path is represented by a segment list or a
set of segment lists which still could be ODN automated steering based. So I
think it maybe good to mention that the draft is talking specifically about
static sid list.
So now if you have two static sid list with two different node sid to two
different endpoint PEs and so when the policy is instantaneous as we are using
prefix sid with node sid flag we are taking advantage of ECMP pathing. So now
if you have threshold constraint set for both candidate paths in the policy how
would that work as the flows are round Robin hashed across the ECMP path.
Since the path can change dynamically but also each path has different latency,
jitter, bandwidth values, how is the best candidate path picked. Also how to
get visibility into all the ECMP flows which is really what you want.
I think the candidate path selection using thresholds can only be used if the
static sid list is a strict path and not loose ECMP path.
This draft is really great work and I like the concept and would be happy to
collaborate on the draft.
Thanks
Gyan
_______________________________________________
spring mailing list
spring@ietf.org
https://www.ietf.org/mailman/listinfo/spring
