[bess] Genart last call review of draft-ietf-bess-evpn-prefix-advertisement-10

[Elwyn Davies <[e-mail]>]

Reviewer: Elwyn Davies
Review result: Almost Ready

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-bess-evpn-prefix-advertisement-10
Reviewer: Elwyn Davies
Review Date: 2018-04-14
IETF LC End Date: 2018-04-10
IESG Telechat date: 2018-05-10

Summary:  Almost ready.  My main concerns are the lack of a good introduction
and a rather weak definition of the format of the new EVPN route option
(looking back on RFC 7342, I think this could be said about that document
also!).  This is very technical material and  a good introduction would help
readers who are not  already deeply into the Data Center area to understand
what is going on.  Also this document has some remaining vestiges of being
written like an academc paper (some were fixed in the previous revision),
particularly the spurious notion of having 'conclusions' (material actually
deserves to be in the Intro)

Major issues:

None

Minor issues:
Lack of a proper introduction: An introduction should precede the terminology
section and needs to be somewhat clearer about the context (presumably
multi-tenant data centers). A reference to RFC 7365 which describes the data
center model in which the EVPN mechanism is expected to be employed would be
very useful. A somewhat expanded version of the text in s2 would be a good
basis for the introduction. The ss2.1 and 2.2 with a short new header would
become a new section (3) which is the Problem Statement.

s5: Associated with my previous comment... An RFC is not a scientific paper and
does not have 'conclusions'.  On reading s5, it strikes me that this text would
actually make quite a nice part of the introduction, probably interpolated
after the first paragraph of the current s2.

Terminology import: The current s1 contains a number of terms that are actually
defined in RFC 7365 (Data Center, Tenant System, Network Virtualization Edge,
etc). Pointing to RFC 7365 s1.1 would be helpful.

s1, VNI: There is some difference between the usage of VNI in RFC 8365 (Section
5.1), in RFC 7365 (s3.1.2) where it means Virtual Network Instance and in this
draft (... Identifier). This is potentially confusing to the naive reader and
definitely confusing with the usage of VNI in item (4) of s4.1 where the VNI is
the VXLAN Network Identifier. It would be better if VNI meant the same thing in
all this closely related work. Please review all instances of VNI in the draft
to make sure they are talking about the same thing.

s2.1:
>    [RFC7432] is used as the control plane for a Network Virtualization
>    Overlay (NVO3) solution in Data Centers (DC),
The acronym NVO3 is used here as opposed to NVO which is used elsewhere in the
document. NVO3 is used in RFC 7365 to refer to an overlay with an L3 underlay
network. It is not quite clear to me whether this is a typo with EVPN NOT being
an NVO3 example or whether the sentence really ought to refer to RFC 7365 and
explicitly say "Network Virtualization Overlay over Layer 3 tunnels". In any
case you can't use an RFC as a control plane - they don't come with knobs on
;-) ! Suggest: NEW: [RFC7432] describes how a BGP MPLS-based Ethernet VPN
(EVPN) can provide the control plane for a Network Virtualization Overlay [over
Layer 3 Tunnels] (NVO[3]) solution in Data Centers (DC), ENDS If this is a real
NVO3 then probably the NVO3 acronym should be used in the rest of the draft in
place of NVO.

s3.1: The encoding of the IP Prefix Route encoding is both underspecified and
to some extent confusing. [I note that this is, in part, inherited from RFC
7342, where I consider Section 7 to be grossly underspecified.] Specifically: -
Figure 3: Expand RD on first use. - 1st bullet after Fig 3: The usage of RD
appears to be defined on a per route type basis in RFC 7342. Accordingly I
don't think it is sufficient to refer to how it is done in RFC 7342. - 2nd
bullet after Fig 3: s/byte/octet/ for consistency - 3rd bullet: Encoding not
specified (presumably unsigned integer) - 4th bullet: The alignment of the
prefix bits in the field is not specified (presumably left aligned). The second
sentence about the size of the field is unnecessary and confusing if the total
length specification is given clearly. - 6th bullet: The alighment/encoding of
the field is not specified (high order doesn't have any meaning without this).
- 7th bullet: It would be better to have the length specification as the first
bullet - this then clarifies the length possibilities of the IP Prefix and GW
IP address fields. Given that the field lengths are given in octets it would be
clearer to specify the total length of the BGP EVPN NLRI variable portion in
octets (and to repeat in s3 as in RFC 7342 that the length is the length of the
varible portion in octets). Thus the length specification would become: NEW: o
The Length field of the BGP EVPN NLRI for an EVPN IP Prefix route MUST be
either 34 (if IPv4 addresses are carried) or 58 (if IPv6 addresses are
carried). The IP Prefix and Gateway IP Address MUST be from the same IP address
family

Nits/editorial comments:
Global: s/i.e. /i.e., /g, s/e.g. /e.g., /g

Global: Section cross references: s/section/Section/

Global: s/route-target/Route Target/ (c.f. definition in RFC 4364 - it might be
useful to reference RFC 4364 in the Terminology section where Route Target is
mentioned.)

Abstract: s/EVPN/The BGP MPLS-based Ethernet VPN (EVPN - RFC 7432) mechanism/

Abstract: Expand NVO on first use and point to RFC 7365.

s1: Mention that many of these terms are fully defined in RFC 7365.

s1, RT-2: Add reference to RFC 7432 Section 7.

s1, RT-5: Note that this is defined in this draft and point to s3.

s1, MAC-VRF and IP-VRF: The terminology definitions define these as tables, but
the usage mostly treats them as entities. For example in the BD terminology
defintion; later in para 1 of s2 we have "IP-VRF tables" - which would mean a
"route table table" if the terminology definition is taken as correct. I think
they need to be defined as entities and a check of all usages made to ensure
that the wording reflects this.

s1, GARP: Add an illustrative reference to RFC 5227. Arguably since there isn't
a separate GARP protocol and there is only one usage, it might be better to
remove this and expand the usage in s4.2.

s1: The terms VXLAN, nvGRE and VTEP need to be defined. Also Ethernet A-D route
(see first comment on s3 below).

s2, last para: This document doesn't provide justification - it wouldn't exist
if the new route type wasn't justified. Suggest: OLD: Once the need for a new
EVPN route type is justified, sections 3, 4 and 5 will describe this route type
and how it is used in some specific use cases. NEW: Then Section 3 describes
the format of an additional option for the BGP EVPN NLRI (see [RFC7432] Section
7) used to carry advertisements of routes using an IP prefix and introduces the
concept of an Overlay Index, describing how it is used with recursive routing
resolution to control the egress path associated with a given IP prefix.
Section 4 describes a set of use cases where the Overlay Index mechanism solves
certain problems encountered in multi-tenant Data Center implementations.
Section 5 summarises the distinction between the single IP address EVPN route
type (RT-2, defined in [RFC742]) and the IP Prefix route type defined in this
document. ENDS

s2.1. para 1: Expand TOR acronym (probably needs a terminology entry).

s2.1, para 2:
> If the term Tenant System (TS) is used to designate a physical or
>    virtual system
What else would it designate (given the definition in RFC 7365)?

s2.1, 3rd bullet after Fig 1: s/depending on who the master is./depending on
whichsystem is the master./

s2.1, para after bullets after Fig 1: Expand ESI on first appearance.

a2.1, last two paras: These contain RFC 2119 keywords (MUST/MAY respectively)
which cannot be used in a requirements outline. Suggest replacing them with
"need to"/"could" respectively.

s2.2, para 2: s/section 2.1/Section 2.1/

s2.2, para 3: s/E.g.:/For example,/, s/1k/1000/ (2 places).

s2.2, next to last para: Need to expand NLRI on first occurrence.

s3, para 1: s/The current/The/ (It won't change in RFC 7432 by definition of
RFCs).

s3, after Figure 2:
>    Where the route type field can contain one of the following specific
>    values (refer to the IANA "EVPN Route Types" registry):
>
>    + 1 - Ethernet Auto-Discovery (A-D) route
>
>    + 2 - MAC/IP advertisement route
>
>    + 3 - Inclusive Multicast Route
>
>    + 4 - Ethernet Segment Route
This is not future proof {and it probably won't be accurate by the time this
draft becomes an RFC) and there is no value in repeating it here. Please delete
it. Note that this removes the expansion of A-D, so will need to add Ethernet
A-D route to Terminology.

s3, after Figure 2:
OLD:
This document defines an additional route type that IANA has added to
the registry, and will be used for the advertisement of IP Prefixes:

+ 5 - IP Prefix Route
NEW:
This document defines an additional route type (RT-5) in the IANA EVPN Route
Types registry [EVPNRouteTypes], to be used for the advertisement of EVPN
routes using IP Prefixes:

Value: 5
Description: IP Prefix Route
ENDS
The reference [EVPNRouteTypes] points to https://www.iana.org/assignments/evpn

s3.1, last para: s/Router's/router's/

s3.1, extra piece: There are various constraints on the values of fields
in the RT-5 variable data:
- The limitations on the value of IP Prefix length and the total length of the
data depending on the address family of IP Prefix and GW IP Address. - The
possible values of the fields depending on the Overlay Index type as set out in
Table 1. The behaviour of receivers if an RT-5 route that breaks these
constraints is received needs to be defined (error behaviour). This might be a
useful point to emphasise that certain data combinations would imply a
withdrawal of the route rather than an advertisement as described in the notes
to Table 1 and elsewhere.

s3.2, para 13: s/sectin 2.2/Section 2.2/

s4.1, para 2: The shorthand SN1/24 which I take it implies subnet SN1 using a
24 bit prefix, needs to be explained on first usage since SN1 is not
immediately obviously an IP address. Suggest: s?subnet SN1/24?subnet SN1, which
uses a 24 bit IP prefix (written as SN1/24 in future),?

s4.1, item (3), 1st bullet: VXLAN and VTEP need to be defined (suggested adding
to terminology above).

s4.2, para 1: s/section 2.1 and 2.2/Sections 2.1 and 2.2/

s4.2, last para: s/GARP message/using a gratuitous ARP REPLY message as
explained in [RFC5227]/ (and remove the GARP erminology entry as suggested
above).

s4.3, para 5 and bullet (2): I suspect AD here should be Ethernet A-D. If not
expand AD on first use - it isn't in terminology (unlike AC). But see previous
notes on Ethernet A-D.

s4.3, para 6: s/in a similar way as/in a similar way to/

s4.3, item (5), bullet 1: Where can I find a definition of the 'MAC disposition
model' and 'VNI disposition model'- Google didn't help me. :-)

s4.4, para 7 (et seq): For consistency: s/ip-lookup/IP-lookup/g and
s/mac-lookup/MAC-lookup/ (2 places) (and s/ip and mac lookups/IP- and
MAC-lookups/)

s4.4, para 8: Expansion of SBD is not required as it is in Terminology (and
might be confusing).

s4.4, para 9:
>  has a IRB interfaces that
>    connect the SBD to the IP-VRF.
Should this be 'has IRB interafces that connect' or 'has an IRB interface that
connects'? If the plural is meant then in the next sentence s/The IRB
interface's/Each IRB interface's/ (I assume the IP/MAC addresses would be
different for each IRB but can't be sure).

s4.4.1, bullets (c) and (d): s/layer-3/layer 3/ (and two further instances in
s4.4.2 and s4.4.3).

s4.4.1, last para: s/like in/as in/

s4.4.3, bullet (c):
OLD:
and there is a requirement to save IP addresses on those interfaces.
NEW:
and there is a requirement to limit the number of IP addresses used.
END

s4.4.3, last para: s/Interface-ful with SBD IRB model/Interface-ful with
unnumbered SBD IRB model/. PS This paragraph does not appear to add value - I
assume it is there to some extent for symmetry with ss4.4.1.and 4.4.2.

s6, para 3: s/any action/any actions/ or maybe, s/any action that end up/any
action that ends up/

s7: This should be redrafted as a request for allocation rather than a report
of a previous action. The current allocation is temporary.

s8.1: idnits reports that EVPN-OVERLAY is now RFC 8365.