Hi Elwyn,
On 2/15/18, 2:12 PM, "Elwyn Davies" <elwynd@xxxxxxxxxxxxxx> wrote:
Reviewer: Elwyn Davies
Review result: Ready with Nits
I am the assigned Gen-ART reviewer for this draft. The General Area
Review Team (Gen-ART) reviews all IETF documents being processed
by the IESG for the IETF Chair. Please treat these comments just
like any other last call comments.
For more information, please see the FAQ at
<https://trac.ietf.org/trac/gen/wiki/GenArtfaq>.
Document: draft-ietf-rtgwg-backoff-algo-07.txt
Reviewer: Elwyn Davies
Review Date: 2018/02/15
IETF LC End Date: 2018/02/14
IESG Telechat date: 2016/02/22
Summary: Ready with nits. The draft does not refer to OSPFv3 - i am not sure if
this is an oversight or because ODSPFv3 already has this mechanism - either way
it should be mentioned. One question that occurred to me is whether the draft
could be considered as updating the OSPFv2/v3 and ISIS standards (not that IETF
has any control over ISIS).
Major issues:
None
Minor issues:
(Non-)Relation between HOLDDOWN_INTERVAL and *_SPF_DELAY values: I notced that
Benjamin Kaduk's SECDIR review of this document
(https://datatracker.ietf.org/doc/review-ietf-rtgwg-backoff-algo-07-secdir-lc-kaduk-2018-02-14/)
was concerned that certain state transitions would never occur. I loooked at
this and realized that his assumption that LONG_SPF_DELAY < HOLDDOWN_INTERVAL
is not required by the document and s6 explicitly resiles from offering
suggested default values. Without this assumption, the state machine appears
to be correct. Not being familiar with the consequences of setting the
HOLDDOWN_INTERVAL relative to the *_SPF_DELAY, I am not sure if anything could
be said about such consequences, but I think it would avoid other people making
the same assumption as the SECDIR reviewer if it was explicitly stated that
HOLDDOWN_INTERVAL is not necessarily bigger than any of the *_SPF_DELAY values
and adding any advice from experience about how to choose appropriate values.
This might also avoid naive implementers shortcutting the state machine
implementation if they made the same assumption.
The definition of HOLDDOWN_INTERVAL explicitly states:
HOLDDOWN_INTERVAL: The time required with no received IGP events
before considering the IGP to be stable again and allowing the
SPF_DELAY to be restored to INITIAL_SPF_DELAY. e.g., 3 seconds. The
HOLDDOWN_INTERVAL MUST be defaulted or configured to be longer than
the TIME_TO_LEARN_INTERVAL.
Perhaps, it should be restated in the third paragraph of section 6.
Requirements Language: Suggest s/RFC2119/RFC8174/ as there are uses of lower
case versions of the reserved words.
I believe this was brought up before and we will make this change. If not, we'll change to the RFC 8174 language.
Default values for parameters: There is a possible conflict between s3, where
example values for the various interval parameters are given and s6 which
states that no default values are specified in the document. The difference in
termnology maybe too subtle for some implementers.
I would expect those implementing IGPs to know the different between an "example" and a "default".
Aborting or otherwise of SPF calculation if an IGP event occurs while an SPF
calculation is in progress. A note about whether this should happen (if it is
possible) would be desirable.
This is certainly out scope and I'm not sure why anyone would deduce from this draft that an implementation should or shouldn't do this.
OSPFv3: Does this (not) equally apply to OSPF v3 for IPv6? If so it should be
mentioned and RFC 5340 included in the references.
Yes. This will be added as reference.
s12: I suspect (although it could be arguable) that the ISIS definition, RFC
2328 (OSPFv2) and (if added) RFC5340 are normative as you need to understand
how they work. This work could even be considered to update these documents.
While implemented from the beginning, SPF Backoff is not specified by the IGP protocol specifications.
Bruno and I will look at the editorial comments and let you know which ones we do and don't incorporate.
Thanks,
Acee
Nits/editorial comments:
General: The term 'back-off' may not be familiar to non-Emglish mother tongue
speakers and on first occurrence needs a little explanation for naive readers
to indicate what it means and to what the back-off is being applied. I have
suggested some additional text to this end for the abstract and s1.
Abstract:
OLD:
This document defines a standard algorithm to back-off link-state IGP
Shortest Path First (SPF) computations.
NEW:
This document defines a standard algorithm to temporararily postpone or
'back-off' link-state IGP Shortest Path First (SPF) computations to reduce
the computational load on IGP nodes if network events occurring at closely
spaced times would otherwise lead to multiple, essentially redundant
recalculations of the routing tables.
ENDS
s1, para 1: s/at the same time/essentially at the same time/
s1, para 2: s/new Shortest Path First (SPF)/new Shortest Path First (SPF)
routing table/
s1, para 2:
OLD:
experiencing multiple temporally close failures over a short
period of time
NEW:
experiencing multiple temporally close failures (that is, eventuating over a
short period of time)
ENDS
s1, para 2: There is a right bracket missing in the following and starting a
clause with 'such as' and ending it with an ellipsis ('...') is redundant. >
such as LDP [RFC5036], RSVP-TE [RFC3209], > BGP [RFC4271], Fast ReRoute
computations (e.g. Loop Free Alternates > (LFA) [RFC5286], FIB updates...
It is unclear to me where the bracket should go: maybe after [RFC5286] or at
the end. Please clarify.
s1, para 2: the phrase
> This also reduces the churn on
> routers and in the network and.
is useless, vague jargon. The previous sentence expresses what I suspect is
meant by 'churn'. so this is redundant and can be omitted.
s1, para 3:
OLD:
To allow for this, IGPs implement an SPF back-off algorithm.
NEW:
To allow for this, IGPs usually implement an SPF back-off algorithm that
postpones or backs-off the running of the SPF calculation when the algorithm
predicts that a run would be essentially redundant or even counter-productive
because it appears that multiple closely timed routing-affecting events can be
expected. ENDS
s1, para 3: s/choosen/chosen/
s2, last bullet: SPF_DELAY is not defined at this point:
s/SPF_DELAY timers values/values for any timers used to back-off SPF
calculations/
s2, last bullet: s/Even though/This is important even though/
s3, para 1: Undesirable ellipsis:
s/a metric change on a link or prefix.../and a metric change on a link or
prefix./
s3:Need to expand SRLG on first use - it isn't deemed to be well-known.
s3, INITIAL_SPF_DELAY bullet: s/A very small delay to quickly handle link
failure/A very small delay to quickly handle a single isolated link failure/
s3, SHORT_SPF_DELAY bullet:
OLD:
SHORT_SPF_DELAY: A small delay to have a fast convergence in case of
a single failure (node, SRLG..), e.g., 50-100 milliseconds.
NEW:
SHORT_SPF_DELAY: A small delay to provide fast convergence in the case of
a single component failure (node, SRLG..) that leads to multiple IGP events,
e.g., 50-100 milliseconds.
ENDS
s5/s5.1: There is currently no text in s5: this is generally considered
inappropriate. Suggest removing the first sentence in s5.1 ("This section
describes the state machine.") and adding to s5: NEW: This section describes
the abstract finite state machine (FSM) intended to control the timing of the
running of SPF calculations in response to IGP events.
s5.1, QUIET bullet: s/occured/occurred/
s5.2: There is no need for 3 expansions of FSM - the expansion can be moved to
s5 as suggested above.
s5.3 title: s/States/State/
s6, next to last para: s/it's RECOMMENDED to play it safe/it is recommended
that timer intervals should be chosen conservatively/ (this is an operational
recommendation).
s6, last para: s/RECOMMENDED/recommended/ (ditto).
s7, para 1: s/is based on/is dependent on/, s/RECOMMENDED/recommended/
(operational again)
s8: Other documents (e.g., from vendors) have used the terms SPF wait time and
SPF hold time. It might be useful to mention that this document essentially
provides ways to implement these settings.
