Hi Alvaro,
thanks for your comments.
Let's first close the ISIS ELC draft before starting to work on OSPF
one, as many comments are common and will be applicable to both ISIS and
OSPF variants.
Please see inline (##PP):
On 29/02/2020 06:00, Alvaro Retana wrote:
Dear authors:
This is my review of draft-ietf-isis-mpls-elc-10. I reviewed this
document alongside draft-ietf-ospf-mpls-elc-12, so many comments are
the same/similar. Thank you for the work in both of them!
Besides the in-line comments, I want to point out here that this
specification is incomplete. It needs to have (1) a formal
description of the new MSD-Type (similar to §5/rfc8491), and (2) a
discussion of the interaction with the BMI-MSD.
I will progress both documents together, so I will wait for both of
them to address my comments before starting their IETF LC.
Thanks!!
Alvaro.
[Line numbers from idnits.]
...
18 Abstract
20 Multiprotocol Label Switching (MPLS) has defined a mechanism to load-
21 balance traffic flows using Entropy Labels (EL). An ingress Label
22 Switching Router (LSR) cannot insert ELs for packets going into a
23 given Label Switched Path (LSP) unless an egress LSR has indicated
24 via signaling that it has the capability to process ELs, referred to
25 as Entropy Label Capability (ELC), on that tunnel. In addition, it
26 would be useful for ingress LSRs to know each LSR's capability for
27 reading the maximum label stack depth and performing EL-based load-
28 balancing, referred to as Entropy Readable Label Depth (ERLD). This
29 document defines a mechanism to signal these two capabilities using
30 IS-IS. These mechanisms are particularly useful, where label
31 advertisements are done via protocols like IS-IS.
[nit] s/ /as the Entropy Label Capability
##PP
done
[minor] "protocols like IS-IS" That last sentence sounds as if there
were other options; for example advertise labels with OSPF and then
use the extensions here. It's just a minor point, but I think that
maybe that last sentence is not needed.
##PP
removed last sentence
...
81 1. Introduction
83 [RFC6790] describes a method to load-balance Multiprotocol Label
84 Switching (MPLS) traffic flows using Entropy Labels (EL). "The Use
85 of Entropy Labels in MPLS Forwarding" [RFC6790] introduces the
86 concept of Entropy Label Capability (ELC) and defines the signalings
87 of this capability via MPLS signaling protocols. Recently,
88 mechanisms have been defined to signal labels via link-state Interior
89 Gateway Protocols (IGP) such as IS-IS
90 [I-D.ietf-isis-segment-routing-extensions]. In such scenarios, the
91 defined signaling mechanisms are inadequate. This draft defines a
92 mechanism to signal the ELC using IS-IS. This mechanism is useful
93 when the label advertisement is also done via IS-IS.
[nit] s/"The Use of Entropy Labels in MPLS Forwarding" [RFC6790]/It also
##PP
done
[nit] s/signalings/signaling
##PP
done
[nit] "In such scenarios, the defined signaling mechanisms are
inadequate." Take this sentence out: the rest of the paragraph is
enough.
##PP
done
95 In addition, in the cases where LSPs are used for whatever reasons
96 (e.g., SR-MPLS [I-D.ietf-spring-segment-routing-mpls]), it would be
97 useful for ingress LSRs to know each intermediate LSR's capability of
98 reading the maximum label stack depth and performing EL-based load-
99 balancing. This capability, referred to as Entropy Readable Label
100 Depth (ERLD) as defined in [I-D.ietf-mpls-spring-entropy-label] may
101 be used by ingress LSRs to determine the position of the EL label in
102 the stack, and whether it's necessary to insert multiple ELs at
103 different positions in the label stack.
[nit] s/in the cases where LSPs are used for whatever reasons/in cases
where LSPs are used
##PP
done
105 2. Terminology
107 This memo makes use of the terms defined in [RFC6790], [RFC4971] and
108 [I-D.ietf-mpls-spring-entropy-label].
[minor] I'm not sure why rfc4971 is referenced here; what terminology
is needed from it?
##PP
removed the reference to rfc4971
...
116 3. Advertising ELC Using IS-IS
118 Even though ELC is a property of the node, in some cases it is
119 advantageous to associate and advertise the ELC with a prefix. In a
120 multi-area network, routers may not know the identity of the prefix
121 originator in a remote area, or may not know the capabilities of such
122 originator. Similarly in a multi-domain network, the identity of the
123 prefix originator and its capabilities may not be known to the
124 ingress LSR.
[minor] Is there a difference that are you trying to highlight between
multi-area and multi-domain? The last two sentences seem redundant to
me; using "domain" should be enough.
##PP
Multi-area and multi-domain are two different cases. I believe it is
important to keep both in the text.
126 One bit of the "Bit Values for Prefix Attribute Flags Sub-TLV"
127 registry defined in [RFC7794] (Bit 3 is desired) is to be assigned by
128 the IANA for the ELC. If a router has multiple line cards, the
129 router MUST NOT announce the ELC for any prefixes that are locally
130 attached unless all of its line-cards are capable of processing ELs.
131 If a router supports ELs on all of its line-cards, it SHOULD set the
132 ELC for every local host prefix it advertises in IS-IS.
[major] The first sentence is not needed because IANA has already
assigned the bit, and any requests should be in the IANA
Considerations section. Perhaps change to something like:
Bit 3 in the Prefix Attribute Flags [RFC7794] is used as the ECL Flag
(E-flag), as shown in Figure 1.
##PP
done
[major] From a general router architecture point of view, I understand
what you mean by line-card. But, strictly speaking from a
specification point of view, what is a line-card? Would using
"interface" instead be an acceptable generalization?
##PP
change to interface
[minor] Is there a difference between "prefixes that are locally
attached" and a "local host prefix"? Are all locally-attached
prefixes host prefixes (/32 or /128)?
##PP
changed to "local host prefix"
[major] "it SHOULD set the ELC for every local host prefix" If ELs
are supported in all the interfaces, when would a router not set the
ELC? IOW, why is "MUST" not used instead of "SHOULD"?
##PP
advertising ELC is not a MUST. It's an optional information that the
originator should advertise, but if it is not, it is not going to break
anything really.
[major/related] The last two sentences seem to be redundant -- I think
that only the second one is needed; suggestion (assuming my
interpretation of the questions above):
##PP
if we only keep the second sentence, the reader may not know what to do
in case where some interfaces do not support ELC. I believe it is good
to keep both.
If a router supports ELs on all of its interfaces, it MUST set the E-flag
(ELC Flag) for every local prefix it advertises.
134 0 1 2 3 4 5 6 7...
135 +-+-+-+-+-+-+-+-+...
136 |X|R|N|E| ...
137 +-+-+-+-+-+-+-+-+...
138 Figure 1: Prefix Attribute Flags
140 E-flag: ELC Flag (Bit 3)
141 Set for local host prefix of the originating node
142 if it supports ELC.
[nit] Justify the description in line with the text (move it to the left).
##PP
done
144 When a router leaks a prefix between two levels (upwards or
145 downwards), it MUST preserve the ELC signaling for this prefix.
[nit] Going up is not really leaking. ;-)
##PP
ok, changed to "propagate".
[minor] An Informative reference to rfc5302 would be nice.
##PP
added.
Suggestion>
When a router distributes a prefix between two levels [RFC5302] it MUST
preserve the E-flag setting.
##PP
I prefer:
"When a router propagates a prefix between ISIS levels [RFC5302], it
MUST preserve the ELC signaling for this prefix."
147 When redistributing a prefix between two IS-IS protocol instances or
148 redistributing from another protocol to an IS-IS protocol instance, a
149 router SHOULD preserve the ELC signaling for that prefix. The exact
150 mechanism used to exchange ELC between protocol instances running on
151 an ASBR is outside of the scope of this document and is
152 implementation specific.
[minor] s/ELC signaling/ELC setting
##PP
we use "signaling" for the inter-area propagation, I would prefer to
keep the same terminology for the redistribution.
[nit] Please expand ASBR.
##PP
done.
[nit] s/ and is implementation specific./.
##PP
done
154 4. Acknowledgements
[major] Move this section to just before the References.
##PP
done
...
161 5. Advertising ERLD Using IS-IS
[major] draft-ietf-mpls-spring-entropy-label says that "To advertise
an ERLD value, a SPRING router: MUST be entropy label capable". This
requirement must be translated to this document so that the ERLD is
only advertised if the ELC is also advertised. I'm assuming that the
ERLD should be ignored if the ELC is not advertised -- but that should
be explicitly defined as well. If the ELC is advertised, but the ERLD
isn't, what value should be assumed, 0?
##PP
RFC8662 already set the rules on when the ERLD can be advertised and
that behavior is orthogonal to protocol which is being used to advertise
the ERLD/ELC.
Same in terms of what should be assumed when the ELC is advertised, but
ERLD is not - has nothing to do with the advertising protocol - should
be specified in RFC8662
I don't feel any of the above should be specified in the IGP ELC drafts.
163 A new MSD-type of the Node MSD ((Maximum SID Depth) sub-TLV
164 [RFC8491], called ERLD is defined to advertise the ERLD of a given
165 router. As shown in Figure 2, it is formatted as described in
166 [RFC8491] with a new MSD-Type code to be assigned by IANA (the type
167 code of 2 is desired) and the Value field is set to the ERLD in the
168 range between 0 to 255. The scope of the advertisement depends on
169 the application. If a router has multiple line-cards with different
170 capabilities of reading the maximum label stack depth, the router
171 MUST advertise the smallest one.
[minor] "new MSD-type...called ERLD is defined to advertise the ERLD"
I suggest that you call the new MSD ERLD-MSD, to differentiate ERLD
from ERLD. ;-)
##PP
changed to:
"A new MSD-type <xref target="RFC8491"/>, called ERLD is defined to
advertise the ERLD of a given router"
[major] s/a new MSD-Type code to be assigned by IANA (the type code of
2 is desired)/the MSD-Type set to 2
IANA already assigned.
##PP
done
[minor] s/Value/MSD-Value
##PP
done.
...
180 When the ERLD MSD-Type is received in the Link MSD Sub-TLV, it MUST
181 be ignored.
[nit] s/When the/If the
##PP
done
183 6. Signaling ELC and ERLD in BGP-LS
...
188 The ELC Flag included in the Prefix Attribute Flags sub-TLV, as
189 defined in Section 3, is advertised using the Prefix Attribute Flags
190 TLV (TLV 1170) of the BGP-LS IPv4/IPv6 Prefix NLRI Attribute as
191 defined in section 2.3.2 of
192 [I-D.ietf-idr-bgp-ls-segment-routing-ext].
[nit] Suggestion>
The ELC is advertised using the Prefix Attribute Flags TLV as defined in
[I-D.ietf-idr-bgp-ls-segment-routing-ext].
##PP
done
194 The ERLD MSD-type introduced for IS-IS in Section 5 is advertised
195 using the Node MSD TLV (TLV 266) of the BGP-LS Node NLRI Attribute as
196 defined in section 3 of [I-D.ietf-idr-bgp-ls-segment-routing-msd].
[nit] Suggestion>
The ERLD-MSD is advertised using the Node MSD TLV as defined in
[I-D.ietf-idr-bgp-ls-segment-routing-msd].
##PP
done
198 7. IANA Considerations
200 IANA is requested to allocate the E-bit (bit position 3 is desired)
201 from the "Bit Values for Prefix Attribute Flags Sub-TLV" registry.
203 IANA is requested to allocate a MSD type (the type code of 2 is
204 desired) from the "IGP MSD Types" registry for ERLD.
[major] IANA has already assigned the values. Suggestion>
Early allocation has been done by IANA for this document as follows:
- Bit 3 in the Bit Values for Prefix Attribute Flags Sub-TLV registry has
been assigned to the ELC Flag. IANA is asked to update the registry to
reflect the name used in this document: ECL Flag (E-flag).
- Type 2 in the IGP MSD-Types registry has been assigned for the ERLD-MSD.
IANA is asked to update the registry to reflect the name used in this
document: ERLD-MSD.
##PP
done
206 8. Security Considerations
208 The security considerations as described in [RFC4971] nd
209 [I-D.ietf-mpls-spring-entropy-label] are applicable to this document.
[minor] Why? Also, I think that some of the other references should
be added here. Suggestion:
This document specifies the ability to advertise additional node
capabilities using IS-IS and BGP-LS. As such, the security considerations
as described in [RFC4971], [RFC7752], [RFC7794], [RFC8491],
[I-D.ietf-idr-bgp-ls-segment-routing-ext],
[I-D.ietf-idr-bgp-ls-segment-routing-msd] and
[I-D.ietf-mpls-spring-entropy-label] are applicable to this document.
##PP
done
211 Incorrectly setting the E flag (ELC capable) (during origination,
212 leaking or redistribution) may lead to black-holing of the traffic on
213 the egress node.
[minor] s/E flag (ELC capable)/E flag
##PP
done
[minor] s/during origination, leaking or redistribution/during
origination or redistribution
##PP
change to:
"during origination, propagation or redistribution"
[major] "...may lead to black-holing of the traffic on the egress
node." I'm not sure I understand how, but the ELC advertisement
should be accompanied by the ERLD-MSD -- see my questions at the
beginning of §5.
##PP
the text talks about the ELC flag only, not about ERLD. If the egress
sets the E flag, while it is not capable of ELC processing the traffic
could be dropped at egress PE - that's what the text says and it is correct.
It is not mandatory to advertise ERLD if the node is ELC capable. It is
however not allowed to advertise ERLD if the node is not ELC capable by
RFC8662.
215 Incorrectly setting of the ERLD value may lead to poor load-balancing
216 of the traffic.
[minor] "may lead to poor load-balancing" If the ERLD is low, then
the traffic may not be load balanced at all...that is not "poor", it
is "0".
##PP
would "lead to poor or no load-balancing" be good enough?
...
244 10.1. Normative References
...
259 [I-D.ietf-mpls-spring-entropy-label]
260 Kini, S., Kompella, K., Sivabalan, S., Litkowski, S.,
261 Shakir, R., and J. Tantsura, "Entropy label for SPRING
262 tunnels", draft-ietf-mpls-spring-entropy-label-12 (work in
263 progress), July 2018.
== Outdated reference: draft-ietf-mpls-spring-entropy-label has been
published as RFC 8662
##PP
done
265 [I-D.ietf-spring-segment-routing-mpls]
266 Bashandy, A., Filsfils, C., Previdi, S., Decraene, B.,
267 Litkowski, S., and R. Shakir, "Segment Routing with MPLS
268 data plane", draft-ietf-spring-segment-routing-mpls-22
269 (work in progress), May 2019.
== Outdated reference: draft-ietf-spring-segment-routing-mpls has been
published as RFC 8660
##done
[minor] This reference can be Informative.
##PP
done
...
276 [RFC4971] Vasseur, JP., Ed., Shen, N., Ed., and R. Aggarwal, Ed.,
277 "Intermediate System to Intermediate System (IS-IS)
278 Extensions for Advertising Router Information", RFC 4971,
279 DOI 10.17487/RFC4971, July 2007,
280 <https://www.rfc-editor.org/info/rfc4971>.
** Obsolete normative reference: RFC 4971
** Replace with rfc7981
##PP
done
...
307 10.2. Informative References
309 [I-D.ietf-isis-segment-routing-extensions]
310 Previdi, S., Ginsberg, L., Filsfils, C., Bashandy, A.,
311 Gredler, H., and B. Decraene, "IS-IS Extensions for
312 Segment Routing", draft-ietf-isis-segment-routing-
313 extensions-25 (work in progress), May 2019.
== Outdated reference: draft-ietf-isis-segment-routing-extensions has been
published as RFC 8667
##PP
done
thanks,
Peter
_______________________________________________
Lsr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/lsr
