Hi Bruno,
:) Lets finalize on the rephrased statement which need to be
updated.
Please see inline [Anil]
Thanks & Regards
Anil S N
"Be liberal in what you accept, and conservative in what you send" - Jon Postel
From: [email protected] [mailto:[email protected]]
Sent: 02 April 2015 18:24
To: Anil Kumar S N (VRP Network BL)
Cc: [email protected]; Alia Atlas; Robert Kebler;
[email protected]; [email protected];
[email protected]; [email protected]; [email protected]; [email protected];
Kalyankumar Asangi; Rajeshmv; Veerendranatha Reddy Vallem; Chris Bowers
Subject: RE: draft-ietf-rtgwg-mrt-frr-architecture-05.txt
Hi Anil,
Please see inline [Bruno]
From: Anil Kumar S N (VRP Network BL) [mailto:[email protected]]
Sent: Thursday, April 02, 2015 10:53 AM
To: DECRAENE Bruno IMT/OLN; Chris Bowers
Cc: [email protected]<mailto:[email protected]>; Alia Atlas;
Robert Kebler;
[email protected]<mailto:[email protected]>;
[email protected]<mailto:[email protected]>;
[email protected]<mailto:[email protected]>;
[email protected]<mailto:[email protected]>; [email protected]<mailto:[email protected]>;
[email protected]<mailto:[email protected]>; Kalyankumar Asangi;
Rajeshmv; Veerendranatha Reddy Vallem
Subject: RE: draft-ietf-rtgwg-mrt-frr-architecture-05.txt
Hi Bruno,Chris,
Lets not talk about number of labels. With respect to TI LFA, In our
implementation we have extended number labels much more than two,
[Bruno] Good job.
- For link protection only, in network with symmetric link cost, up to 2 labels
may be required for. Never more than 2.
- For others cases, in particular node and/or SRLG protection, more than 2
labels may be required. In theory, there is no limit. In practice, there is a
limit. In our networks, the max found was 4 labels. So it's good to be able to
push more than 2 labels to enforce the FRR detour.
[Anil] : Agreed
Ex: when there is no PQ node also when there no adjacency b/w any P and Q node
pair. We find a shortest path from P & Q form; a label stack that will be
pushed. In case L3 VPN PE label stack could become much more. This consumes
more computation.
Best effort label for a destination is Destination Label , SR FRR label for the
same destination is [P Node -> Label stack to reach Q Node -> Destination
label].
[Bruno] I would propose
OLD: using an MPLS label stack with two labels
NEW: by pushing up to two additional labels
There is a possibility of Looping in TI-LFA; We can expect in case of multiple
simultanious failure:
[Bruno] You are right. In case of independent/non FRR planned failures, the
first PLR can't avoid all failures since those failures were expected to not
planned/expected to occur at the same time. The second PLR will try to protect
the second failure. In good cases, the two protections are complementary and
will achieve "de facto" protection. In bad cases, forwarding loop will occur.
In theory, the handling of the second independent failure, is a (difficult)
choice, between dropping or trying to repair. In theory TI-LFA could do both,
but as of today, it tries the second protection
This is not specific to TI-LFA. FRR assumes _local_ protection (hence local
knowledge) of a _pre_computed_ (hence pre-planned) failure.
I have been told ARC could handle multiple consecutive FRR (through some
localize notification mechanism) but I don't have the details.
[Anil] : I also suggested authors to add comparison with respect to ARC too.
Consider the case where packet is been forwarded towards destination from a
source on backup FRR path with TI LFA backup path Statck as below :
P Node Lable
P-Node to Q-Node Adj lable
Q-Node Lable
Destination Lable
Topology :
P----Q (Cost 10)
Q----Destination(Cost 10)
P----Destination (Cost 100)
Faliure happens on the only link b/w Q-Node towards destination; Reachability
to destination with SPF calculation or with FRR falls back towards P-Node (Q's
TI LFA calculation).
[Bruno] In this specific case, I don't see a loop. Q, acting as the second PLR
(in a row), has computed that node "P" is its P node and node "D" is its Q
node. P is reached via a node-SID (possibly implicit label) which is not
affected by the first failure. Q is reached by a Adj-sid, which is not affected
by the first failure.
[Anil] : Here P would send a packet to Q and as primary path to D is lost Q
would return packet back to P till P learns link b/w Q and D is down.
So there could be a loop.
[Bruno] I could not find one in this example, but I agree that we could find
examples with loops.
e.g. the following topology
A - B
| \ |
C - D
Where all links have a metric of 10, except AC which has 100. For the traffic
from A to C, if both AC and BD links fails at the same time (and were not
identified as SRLG), there will be a loop between A and B. (and funnily (is it?
;-) ) as packets are looped, the label stack seems to increase to "infinity" as
additional P nodes (alternatively C and D) are used.
[Anil] : I understand Loop here but Label stack increase can you help me to
understand.
Here primary path would be A->D->C [Cost 20] for A (LFA A->B->D->C), Primary
path for B to would be B->D->C [Cost 20] (LFA B->A->D->C).
On A->C link failure A chooses A->B->D->C, On B->D failure B chooses
B->A->D->C. A would send packet to B and B would send packet to A.
Here which label is inserted ?
10
A B
+---------+
| \ |
| \ |
| \ |
| \10 | 10
100 | \ |
| \ |
| \ |
| \|
+---------+
C 10 D
So I still feel there is a tradeoff.
[Bruno] There are many tradeoff ;-). But this one is about handling a second
independent failure. Not about increase coverage of a protected resource
(link/node/SRLG/any)
[Anil] : This Could be explicitly specified :)
Thanks
Regards
/Bruno
Thanks & Regards
Anil S N
"Be liberal in what you accept, and conservative in what you send" - Jon Postel
_______________________________________________
rtgwg mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/rtgwg
