The added classification of alternates looks fine, except I am
completely confused about your statement about NP. See inline.
Alia Atlas wrote:
I'm completely confused by this last statement. Surely it is possible
to have a node protecting LFA on a pt-pt interface?
On 10/12/07, Rüdiger A Martin <[email protected]>
Alia, dear Alex,
we have taken a look at your document and would like to add a few
comments to the discussion already going on:
- The concept of node protecting LFAs is first mentioned on page 6. The
definition and explanation follows later.
True - The first place a node protecting LFA is mentioned, I've added a
reference to the section with the definition.
I don't see a clean way of defining it at/before that location.
There are several kinds of loop-free alternates. But the suggested
order of preference among those different LFAs does not become clear at
all in the draft. So there should be some paragraph where all possible
candidate types are listed and described with their pros and cons. That
description should include the protected failure types but also the
behavior in case of unprotected failures: e.g., I miss a clear statement
that downstream LFAs do not create any loops no matter what kind of
failures may happen. Based on this, a clear suggestion for an order of
preference can be given.
Yes, we didn't have consensus about the order of preferences - and it
on what other types of protection are available in the network. I
agree that this
would be useful. How does the following added section (after 3.6)
"3.7. LFA types and Trade-offs
LFAs can provide different amounts of protection and the decision
about which type to prefer is dependent upon network topology and
other techniques in use in the network. This section describes the
different protection levels and the trade-offs associated with each.
1. Primary Next-hop: When there are equal-cost primary next-hops,
using one as an alternate is guaranteed to not cause micro-loops
involving S. Traffic flows across the paths that the network will
converge to, but congestion may be experienced on the primary
paths since traffic is sent across fewer. All primary next-hops
are downstream paths.
2. Downstream Paths: A downstream path, unlike an LFA, is guaranteed
not to cause a micro-loop involving S regardless of the actual
failure detected. However, the expected coverage of such
alternates in a network is expected to be poor. All downstream
paths are LFAs.
3. LFA: An LFA can have good coverage of a network, depending on
topology. However, it is possible to get micro-loops involving S
if a more severe failure occurs than is protected against.
The different types of protection are abbreviated as LP (link-
protecting), NP (node-protecting) and SP (SRLG-protecting).
a. LP, NP, and SP: If such an alternate exists, it gives protection
against all failures.
b. LP and NP: Many networks may handle SRLG failures via another
method or may focus on node and link failures as being more
c. LP: A network may handle node failures via a high availability
technique and be concerned primarily about protecting the more
common link failure case.
d. NP: These only exist on interfaces that aren't point-to-point.
If link protection is handled in a different layer, then an NP
alternate may be acceptable.
I'd welcome suggestions on how to improve this.
Page 7: "Since the funcionality of link and node protecting LFAs is
greater than that of downstream paths, a router SHOULD select a link and
node protecting LFA over a downstream path." - That reads like
downstream paths are generally worse. But of course a downstream link-
and node-protecting LFA is better than a non-downstream link- and
node-protecting LFA. So this point is strongly related to the point
mentioned earlier concerning the order of preference.
Added the clarification of "link-protecting downstream paths".
This is related to the observation (which we already have in section
1.1 "There are topologies where there may be either a node portecting
LFA, a downstream path, both or neither") i.e. node protecting LFAs and
link protecting downstream paths are not strict sub-sets.
Yes. That is a good point to make.
A few comments on the algorithm on page 16 (besides I agree with
* Step 10: This means, if the router prefers other primary next
hops, it prefers them no matter what kind of failures (links or nodes)
they protect. So this algorithm prefers a link-protecting primary over a
link- and node-protecting downstream LFA. (Depending on the order of the
candidates - see below.)
True - this is intended to reflect the behavior for the network once
it has converged.
I could instead embed the preference at Step 16 - but that flexibility
is there already.
* The output of the algorithm depends on the order of the
candidate next hops.
1) Imagine that H_1 is a link-protecting primary next hop.
H_2 is a link- and node-protecting downstream LFA. Then H_1 will be
selected in the first iteration. During the second iteration, step 11
selects H2 instead.
2) Now imagine H_1' = H_2 and H_2' = H_1. Then the first
iteration selects H_1'. However, the second iteration selects H_2' as
the candidate in step 10. So the final result is H1.
I think this problem is due to the missing order of
Ah - nope - it is due to a missing step - added between step 10 &
"If cand_type is not PRIMARY, P_i.alt_type is PRIMARY and the router
prefers other primary next-
hops for use as the alternate, then continue to the next
we really appreciate this document and support the progression
of the document after consideration of the comments from above. We think
that Mike's idea to include explicitly that this may only provide a
partial solution is worth considering. But maybe together with the
statement that this is a really simple and easy to implement approach
since it does not require any support (i.e. signalling) from other
Ok - so at the end of the abstract, I've added
"This simple approach does not require any support from other routers.
extent to which this goal can be met by this specification is
dependent on the topology of the network."
Thanks for the comments and suggestions!
Dipl.-Inform. Rüdiger Martin
Department of Distributed Systems (Informatik III)
Insitute of Computer Science, University of Würzburg
Am Hubland, 97074 Würzburg, Germany
Tel: +49 931 888-6651
Fax: +49 931 888-6632
rtgwg mailing list
rtgwg mailing list
rtgwg mailing list