[Detnet] Benjamin Kaduk's No Objection on draft-ietf-detnet-data-plane-framework-04: (with COMMENT)

[Benjamin Kaduk via Datatracker <[e-mail]>]

Benjamin Kaduk has entered the following ballot position for
draft-ietf-detnet-data-plane-framework-04: No Objection

When responding, please keep the subject line intact and reply to all
email addresses included in the To and CC lines. (Feel free to cut this
introductory paragraph, however.)


Please refer to https://www.ietf.org/iesg/statement/discuss-criteria.html
for more information about IESG DISCUSS and COMMENT positions.


The document, along with other ballot positions, can be found here:
https://datatracker.ietf.org/doc/draft-ietf-detnet-data-plane-framework/



----------------------------------------------------------------------
COMMENT:
----------------------------------------------------------------------

Other than some remarks on the security considerations, I basically just
have a bunch of editorial comments.  I've tried to deduplicate with what was
already reported by other ADs, but no doubt have kept a few things in that
have already been mentioned.

Section 1

   The DetNet Architecture models the DetNet related data plane
   functions decomposed into two sub-layers: a service sub-layer and a
   forwarding sub-layer.  The service sub-layer is used to provide

nit: "models" needs two objects to act on/compare, so maybe "as being
decomposed into".

   replicated in other forwarding technologies.  Most of DetNet benefits
   can be gained by running over a data link layer that has not been
   specifically enhanced to support all TSN capabilities but for certain
   networks and traffic mixes delay and jitter performance may vary due
   to the forwarding sub-layer intrinsic properties.

nit: I think the "certain networks and traffic mixes" are supposed to be
contingent on "a data link layer that [is not a TSN]" but the current text
does not quite match that.  Perhaps "certain such networks" is the smallest
fix, though it does conflate "data link layer" and "network" in an
unfortunate manner.

Section 3

   connectivity function of the forwarding sub-layer.  An example of
   this is Packet Replication, Elimination, and Ordering functions see
   Section 4.3.  The ordering (POF) uses sequence numbers added to

nit: the punctuation around the reference should be tweaked.

   The method of instantiating each of the layers is specific to the
   particular DetNet data plane method, and more than one approach may
   be applicable to a given bearer network type.

What is a "bearer network"?

Section 3.1.1

   applying existing standardized headers and/or encapsulations.  The
   Detnet forwarding sub-layer may provide capabilities leveraging that
   same header or encapsulation technology (e.g., DN IP or DN MPLS) or
   it may be achieved by other technologies (e.g., Figure 2).  DetNet is

nit: Figure 2 is not a technology; maybe "as shown in Figure 2".

Section 3.1.2

   number) in packets.  For example, in DetNet IP, zero encapsulation is
   used and no sequence number is available, and in DetNet MPLS, DetNet
   specific information may be added explicitly to the packets in the
   format of S-label and d-CW [I-D.ietf-detnet-mpls] .

nit: s/format of/form of/

Section 3.3

   Number (for PREOF) for each DetNet flow.  The DetNet Service sub-
   layer requires both; the DetNet forwarding sub-layer requires only
   Flow-ID.  Metadata can also be used for OAM indications and

nit(?): to say that "the service sublayer requires both" to me implies that
it always and unconditionally requires both, but IIUC the PREOF functions
are optional, and so for a given flow the service sublayer might not require
the sequence number.  Perhaps swapping it around to be structured more like
"the flow-ID is used by both the service and forwarding sub-layers, but the
sequence number is only used by the service layer" would help?

   Some MPLS examples of implicit metadata include the sequence number
   for use by the PREOF function, or even all the essential information
   being included into the DetNet over MPLS label stack (the DetNet
   Control Word and the DetNet Service label).

I would have thought that the detnet elements in the label stack would be
considered explicit metadata, but perhaps I'm just looking through the wrong
lens.

Section 3.4

   One method of operating an IP DetNet data plane without encapsulation
   is to use "6-tuple" based flow identification, where "6-tuple" refers
   to information carried in IP and higher layer protocol headers.
   General background on the use of IP headers, and "6-tuples", to
   identify flows and support Quality of Service (QoS) can be found in
   [RFC3670].  [RFC7657] provides useful background on differentiated

I did not see the explicit phrase "6-tuple" used in RFC 3670, so I don't
think this text is quite right as written.  (7657 does talk about it, but is
not currently being referenced for that purpose.)

Section 3.6.1.5

Is there a difference between "packet-by-packet distribution [of the same
flow]" and "packet-by-packet load sharing"?

Section 3.6.1.6

   as those in use by IP and MPLS networks.  At the Application layer a
   client of a DetNet service can use existing techniques to detect and
   monitor delay and loss.

[Is there a good reference for these "existing techniques"?  I recognize
that we shouldn't go into extensive detail here, and absent a single
consolidated reference, no-reference may be best.]

Section 3.6.2

   Service protection allow DetNet services to increase reliability and
   maintain a DetNet Service Assurance in the case of network congestion
   or network failure.  Detnet relies on the underlying technology

My understanding is that the whole point of DetNet was to provide
reliability in the face of network congestion (e.g., by resource
reservation).  So is it really only the "network failure" case that benefits
from service protection?

Section 3.6.3.1

   IP aggregation has both data plane and controller plane aspects.  For
   the data plane, flows may be aggregated for treatment based on shared
   characteristics such as 6-tuple.  Alternatively, an IP encapsulation

In what way would 6-tuple be a "shared characteristic" for *different*
flows?  I thought 6-tuple was typically used as a unique flow identifier...
(I see in Section 4.2.1 we refer to flows that "share 6-tuple attributes".)

Section 4.1

   However, from a practical and implementation standpoint, they are not
   equivalent at all.  Some approaches are more scalable than others in
   terms of signaling load on the network.  Some can take advantage of
   global tracking of resources in the DetNet domain for better overall
   network resource optimization.  Some are more resilient than others
   if link, node, or management equipment failures occur.  While a
   detailed analysis of the control plane alternatives is out of the
   scope of this document, the requirements from this document can be
   used as the basis of a later analysis of the alternatives.

side note: using "some" so much requires the reader to know which are which
... that may be reasonable, but perhaps we want the reader's brainpower
working on other things.

Section 4.2

nit: RestConf and YANG probably are best considered together as a single
"management mechanism" for these purposes.

Section 4.2.2

   DetNet application with the required service.  A requirement for the
   DetNet Controller Plane will be the ability to assign a particular
   identified DetNet IP flow to a path through the DetNet domain that
   has been assigned the required nodal resources.  This provides the

nit: I don't think that "nodal" is quite the right word, here, as it refers
to a property of the (network) graph more than the specific element
colocated with the node in the graph.  "per-node" seems like a better fit,
to me.

Section 4.3

   domain requires explicit support.  There may be capabilities that can
   be used, or extended, for example GMPLS end-to-end recovery [RFC4872]
   and GMPLS segment recovery [RFC4873].

nit: maybe this is "existing capabilities"?

Section 5

I guess there's some standard security considerations for when flow
aggregation is in play, namely that misbehavior from one component flow can
affect sibling flows as collateral damage.

   Security considerations for DetNet are described in detail in
   [I-D.ietf-detnet-security].  General security considerations are

The referenced document (now at -09) seems significantly improved from when
I previously made an in-passing review of the -03
(https://mailarchive.ietf.org/arch/msg/detnet/jZLodXBmQa7ZFDbBIvDO_xCDD0E/),
however, some of the issues I mentioned there remain, at least in part
(e.g., mentioning the use of HMAC without a colocated discussion of the need
for key distribution and having a taxonomy of threats titled "threat
model").  I would like to know what the current maturity state of the
detnet-security document is believed to be, so as to judge how appropriate
it is for the data-plane-framework to refer to it in this manner.

   described in [RFC8655].  This section considers general security

I think this is more like "Architecture-level DetNet" than "Generic", as the
current formulation sounds like it should be referring to BCP 72 :)

   Security aspects which are unique to DetNet are those whose aim is to
   provide the specific quality of service aspects of DetNet, which are
   primarily to deliver data flows with extremely low packet loss rates
   and bounded end-to-end delivery latency.

I can't find a way to read this other than "security aspects [...] whose aim
is to provide the specific quality of service aspects", and I'm puzzling at
why it's phrased in this way.  Specifically, I thought that *providing* the
QoS aspects is the core role of DetNet, and thus that the *security* aspects
would be protecting those aspects in the face of (some class of) attackers.
Would it be appropriate to replace this with something like the following?

% Part of what makes DetNet unique is its ability to provide specific and
% reliable quality of service (delivering data flows with extremely low
% packet loss rates and bounded end-to-end delivery latency), and the
% security-related aspects of protecting that quality of service are
% similarly unique.

This might also be a good place to note that correct operation of the PREOF
functions is pretty critical, and failures there could lead to pretty
catastrophic situations for the operational technology relying on them.  I
don't think there are ways for an attacker to try to attack them directly,
but Murphy's Law can be a pretty effective attacker, too.

   The primary considerations for the data plane is to maintain
   integrity of data and delivery of the associated DetNet service
   traversing the DetNet network.  Application flows can be protected

I'd suggest leading in to this with "As for all communications protocols,"
to be clear that this paragraph is not trying to cover the bits that are
"unique to DetNet" as much as remind the reader of general best practices.

   At the management and control level DetNet flows are identified on a
   per-flow basis, which may provide controller plane attackers with
   additional information about the data flows (when compared to
   controller planes that do not include per-flow identification).  This
   is an inherent property of DetNet which has security implications
   that should be considered when determining if DetNet is a suitable
   technology for any given use case.

I might also note that attackers with access to the controller plan already
have pretty significant attacks available to them.

   To provide uninterrupted availability of the DetNet service,
   provisions can be made against DOS attacks and delay attacks.  To
   protect against DOS attacks, excess traffic due to malicious or
   malfunctioning devices can be prevented or mitigated, for example
   through the use of existing mechanism such as policing and shaping
   applied at the input of a DetNet domain.  To prevent DetNet packets

In my head this would include things like disabling switch ports that are
receiving traffic floods that might overload the switch CPU, in order to
protect the DetNet traffic on other ports.  I don't see much value in saying
that in the document, though, since it's more of an operational practice
thing.

Section 9.1

RFC 3209 is cited only once, and it does not seem like a particularly
normative one.  Similarly, RFC 3473 is cited only as a "such as" example
(though several times).