Robert,

Comments inline

Yours Irrespectively,

John

From: spring [mailto:[email protected]] On Behalf Of Robert Raszuk
Sent: Monday, May 23, 2016 4:14 PM
To: Adrian Farrel
Cc: idr wg; [email protected]
Subject: Re: [spring] [Idr] New draft for data center gateways

Dear Authors,

Question 1:

Assume that prefix X has been advertised with tunnel attribute as described in 
the draft with both GW1 and GW2 entry points to Egress DC site.

So remote GW in Ingress DC site receives at least one such advertisement and as 
each contains both GW1 and GW2 entries it can engineer flows to those.

So far so good ... but what happens when link between PE (on left side) and GW1 
goes down ?

BGP will after some time remove that path via all ASes, but GW2 will keep 
advertsing prefix X as still reachable via both GW1 and GW2 within tunnel 
encapsulation attribute as from his perspective nothing will be wrong.

How remote GW in Ingress DC site is now supposed to know that GW1 link to PE in 
AS2 went down and stop pushing traffic towards it ?


[JD]  We were assuming that a) GW/backbone links would be advertised in BGP LS 
(optionally w/ EPE) and b) a GW that is disconnected from the backbone does not 
advertise an auto-discovery route.  This will be made explicit in the next 
revision.


- - -

Question 2:

What happens if all L3 DC CLOS IP Fabrics use eBGP not IGP ?


[JD]   We make no assumptions about what control plane is used within a DC.  
After all, the draft is about DC “interconnect”.  (Btw, it should be spelled 
‘Clos’ since it’s a person’s last name.)


- - -

Question 3:

Is per prefix tunnel attribute where say /32 routes may be exchanges flat (ref 
calico approach - https://www.projectcalico.org/) really a scalable solution ?


[JD]   Are you simply repeating the opinion you expressed when the 
tunnel-encaps draft first came out, or do you have a scalability concern that 
hasn't already been discussed?


Best,
Robert.

PS. Just purely as information your inter DC traffic steering problem 
description is easily solved already today with one level of indirection - 
Example: LISP. Not sure if we need additional flat protocol extensions here.



_______________________________________________
spring mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/spring

Reply via email to