Hi Xufeng,
Thanks for the comments. Please see inline …
From: CCAMP <ccamp-bounces@xxxxxxxx>
On Behalf Of Xufeng Liu
Sent: 28 January 2020 22:02
To: Rob Wilton (rwilton) <rwilton=40cisco.com@xxxxxxxxxxxxxx>
Cc: Aihua Guo <aihuaguo@xxxxxxxxxxxxx>; draft-ietf-ccamp-layer0-types.all@xxxxxxxx; last-call@xxxxxxxx; yang-doctors@xxxxxxxx; ccamp@xxxxxxxx; Italo Busi <Italo.Busi@xxxxxxxxxx>
Subject: Re: [CCAMP] Use of identityref and enumerations (was RE: Yangdoctors last call review of draft-ietf-ccamp-layer0-types-03)
Hi Italo, YANG Doctors,
To take a couple of concrete examples from the draft (https://tools.ietf.org/html/draft-ietf-ccamp-layer0-types-03):
identity layer0-bandwidth-type
- I can see that different channel signal types might be defined in future, by vendors, or other different SDOs. Having these as identities gives that flexibility to define new names for signal types of bandwidths..
But, identity dwdm-ch-spc-type is somewhat of a different beast.
Its values are for 100Ghz, 50 Ghz, 25 Ghz, 12.5 Ghz channel spacing.
So, what happens if two different vendors both want to define 6.25 Ghz channel spacing? You end up with two different identities for the same value. Clients would need to understand both vendor A's 6.25 Ghz identity and vendor B's 6.25 Ghz identity where
they are really representing the same thing. Hence, my desire to see this an enumeration.
In this case, I would prefer to use identify instead of enumeration. Even if vendor A and vendor B both augment the type with 6.25G, the handling may or may not be the same. Allowing the clients to distinguish
the two independent augmentations could be a benefit. Even if the handlings are the same, the clients could still deal with the situation (consciously).
[RW]
But this seems bad to me, i.e. to have the same numerical values with different meanings!
Do we really want Cisco DWDM with 6.25 Ghz channel spacing to be different from Volta DWDM with 6.25 Ghz channel spacing? So now the client doesn’t need to just know that they want the channel space to be 6.25 Ghz, but they need
to be sure that they express the numerical value in the right way for the device.
If the enumeration type were used, vendor A and vendor B might both deviate the type with separate sets of values, the clients would not be able to take both sets and deal with both vendors.
Depending on the situation, some of the vendor A's values or vendor B's values might never come to IETF to be standardized. Enumeration would make the vendor support more difficult.
If you want increased flexibility, perhaps consider defining it as a union of enum and decimal value.
For the purpose of identity instead of quantity, I would avoid the decimal type, for which RFC7950 only specifies the external string representation. Internally the system may use float or double, which might
introduce inconveniences or issues of encoding differently or losing precisions.
[RW]
I suspect that this wouldn’t be a problem for these types where the required precision is significantly less than a double.
Thanks,
Rob
This could be done either in the standard model or a vendor specific deviation-replace of the type. This would allow is to be configured either using a standard defined enum name, or as a raw numerical value,
where the device would police what numerical values are allowed.
My interpretation is that flexi-ch-spc-type, flexi-slot-width-granularity, and cwdm-ch-spc-type are all like dwdm-ch-spc-type.
Using identities for fec-type is fine.
Do any of the other YANG Doctors have a view on this?
Thanks,
Rob
> -----Original Message-----
> From: Italo Busi <Italo.Busi@xxxxxxxxxx>
> Sent: 28 January 2020 01:28
> To: Rob Wilton (rwilton) <rwilton@xxxxxxxxx>; Zhenghaomian
> <zhenghaomian@xxxxxxxxxx>; Aihua Guo <aihuaguo@xxxxxxxxxxxxx>; yang-
> doctors@xxxxxxxx
> Cc: last-call@xxxxxxxx;
draft-ietf-ccamp-layer0-types.all@xxxxxxxx;
> ccamp@xxxxxxxx
> Subject: Use of identityref and enumerations (was RE: [CCAMP] Yangdoctors
> last call review of draft-ietf-ccamp-layer0-types-03)
>
> Looking at some YANG doctor review comments, I have the feeling that there
> is some debate about whether we should use identityref or enumeration
> types, which requires some broader discussion to understand which
> guidelines we can follow since it is impacting multiple YANG models we are
> working on
>
> The key question is about future extensibility.
>
> I have understood that it is possible to extend an enumeration after the
> data plane standards are updated by revising the YANG model. Therefore
> possible data plane standard extensions are not a criteria for preferring
> identities to enumeration.
>
> However, there are some cases where vendor-private extensions are
> possible.
>
> For technical and political reasons, I would prefer not to add vendor-
> private values to a standard enumeration especially when dealing with data
> plane standards not owned by IETF.
>
> I have some concerns with using "deviate replace" for an enumeration since
> these are more extensions than deviations. Having vendor-private
> identities extending the standard ones seems a more natural choice to
> describe an extension and to allow the client understanding that the
> server supports such an extension.
>
> Moreover, even if two vendors supports similar extensions, there is no
> guaranteed of interoperability between them. The two vendors can define
> their own identities within their own namespace so the clients will not be
> confused.
>
> Therefore, my suggestion is that we should use identities when we think it
> is possible to have vendor-private extensions
>
> Just my 2 cents
>
> Italo
>
> > -----Original Message-----
> > From: Rob Wilton (rwilton) [mailto:rwilton@xxxxxxxxx]
> > Sent: martedì 21 gennaio 2020 14:35
> > To: Zhenghaomian <zhenghaomian@xxxxxxxxxx>; Aihua Guo
> > <aihuaguo@xxxxxxxxxxxxx>;
yang-doctors@xxxxxxxx
> > Cc: last-call@xxxxxxxx;
draft-ietf-ccamp-layer0-types.all@xxxxxxxx;
> > ccamp@xxxxxxxx
> > Subject: Re: [CCAMP] Yangdoctors last call review of
> > draft-ietf-ccamp-layer0-
> > types-03
> >
> > Hi Haomian, Aihua,
> >
> > For "layer0-bandwidth-type" I think that identities are okay. It
> > seems quite plausible that there could be augmentations of other types
> > here. I do wonder whether this is really "layer0-bandwidth-type" or
> > should it be something like "layer0-signal-type"? E.g. I was just
> > looking at the Wikipedia entry for "Optical Transport Network" and it
> refers to these as signals rather than bandwidths:
> >
https://en.wikipedia.org/wiki/Optical_Transport_Network.
> >
> > For slot-width and channel-spacing, I think that enumerations would be
> > better than identities. I.e. if smaller channels are required then I
> > think that it would be better to update this base types YANG module,
> > rather than potentially define them in other YANG modules.
> >
> > If you concern is about vendors needing smaller channels then there
> > are few options to mitigate:
> > (i) define the smaller slots in the enumeration now (if you know they
> > are coming).
> > (ii) vendors can "deviate replace" on the type of the leafs, e.g. to
> > an enumeration with extra values, or a union of an enumeration and a
> > numerical value.
> >
> > Thanks,
> > Rob
> >
> >
> > > -----Original Message-----
> > > From: Zhenghaomian <zhenghaomian@xxxxxxxxxx>
> > > Sent: 17 January 2020 12:33
> > > To: Aihua Guo <aihuaguo@xxxxxxxxxxxxx>; Rob Wilton (rwilton)
> > > <rwilton@xxxxxxxxx>;
yang-doctors@xxxxxxxx
> > > Cc: last-call@xxxxxxxx;
draft-ietf-ccamp-layer0-types.all@xxxxxxxx;
> > > ccamp@xxxxxxxx
> > > Subject: 答复: Yangdoctors last call review of
> > > draft-ietf-ccamp-layer0-
> > > types-03
> > >
> > > Hi, Robert, Aihua,
> > >
> > > Thank you for the feedback, please see inline with [Authors2], we
> > > are getting closer:)
> > >
> > > Best wishes,
> > > Haomian
> > >
> > >
> > > -----邮件原件-----
> > > 发件人: Aihua Guo [mailto:aihuaguo@xxxxxxxxxxxxx]
> > > 发送时间: 2020年1月17日 4:07
> > > 收件人: Rob Wilton (rwilton) <rwilton@xxxxxxxxx>; Zhenghaomian
> > > <zhenghaomian@xxxxxxxxxx>;
yang-doctors@xxxxxxxx
> > > 抄送:
last-call@xxxxxxxx;
draft-ietf-ccamp-layer0-types.all@xxxxxxxx;
> > > ccamp@xxxxxxxx
> > > 主题: RE: Yangdoctors last call review of
> > > draft-ietf-ccamp-layer0-types-03
> > >
> > > Hi Rob, Haomian,
> > >
> > > Re: using (floating point) values in a leaf vs. using identityref
> > > for optical bandwidth, spacing, slot-width etc, I'd lean towards
> > > using the identityref. Two main reasons from my point of view:
> > >
> > > a) in optical layer, the value of bandwidth, slot width, channel
> > > spacing etc. are fixed and discrete, and is always determined by the
> > > technologies used underneath, e.g. FEC or modulation format. Those
> > > values are frequently compared/checked by the program. If using
> > > floating points the value comparison is not definitive, which may
> > > make it difficult for the program logic to handle. No such issue if
> > > using identityref. I'd prefer to avoid encoding those fixed values as
> floating points.
> > >
> > > b) Optical technologies are evolving and new definitions such as new
> > > channel spacing or slot width would mandate changes to be made in
> > > the original model if using the numerical value leaf nodes. It may
> > > also cause backward compatibility issues. With identityref one
> > > creates new definitions in a new and augmented model, rather than
> > > making changes over the original model. YANG validation on the
> > > client side can filter out the new values automatically if it
> > > detects that the new values are not supported by the server (because
> > > the augmented model is not
> > implemented).
> > >
> > > Thanks,
> > > Aihua
> > >
> > > -----Original Message-----
> > > From: Rob Wilton (rwilton) <rwilton@xxxxxxxxx>
> > > Sent: Thursday, January 16, 2020 5:08 AM
> > > To: Zhenghaomian <zhenghaomian@xxxxxxxxxx>;
yang-doctors@xxxxxxxx
> > > Cc: last-call@xxxxxxxx;
draft-ietf-ccamp-layer0-types.all@xxxxxxxx;
> > > ccamp@xxxxxxxx
> > > Subject: RE: Yangdoctors last call review of
> > > draft-ietf-ccamp-layer0-
> > > types-03
> > >
> > > Hi Haomian,
> > >
> > > Sorry, I had missed your email over Christmas. Replying back to the
> > > original email to capture the other aliases.
> > >
> > > Please see inline ...
> > >
> > >
> > >
> > > > -----Original Message-----
> > > > From: last-call <last-call-bounces@xxxxxxxx> On Behalf Of
> > > > Zhenghaomian
> > > > Sent: 25 December 2019 08:20
> > > > To: Rob Wilton (rwilton) <rwilton@xxxxxxxxx>;
> > > > yang-doctors@xxxxxxxx
> > > > Cc: last-call@xxxxxxxx;
> > > >
draft-ietf-ccamp-layer0-types.all@xxxxxxxx;
> > > > ccamp@xxxxxxxx
> > > > Subject: Re: [Last-Call] Yangdoctors last call review of
> > > > draft-ietf-ccamp-
> > > > layer0-types-03
> > > >
> > > > Hi, Robert,
> > > >
> > > > Again, thank you very much for your detailed review and comments.
> > > > The authors had a few iterations of discussions and give the
> > > > update for this document as follow.
> > > >
> > > > Model update can be found as the pull request:
> > > >
https://nam03.safelinks.protection.outlook.com/?url="">
> > > > gi
> > > > th
> > > > ub.com%2Fietf-
> > &data="" href="http://40futurewei.com" target="_blank">40futurewei.com%7C93555aa
> > > > 03
> > > >
> > 041406a21ec08d79a6bf5dc%7C0fee8ff2a3b240189c753a1d5591fedc%7C1%7
> > C1%7
> > > > C6
> > > >
> > 37147660797808234&sdata=JPWe96JXR556EQk%2FSsaCgVf3QuK%2FGl
> > 0ha1eQ
> > > > qS
> > > > NEfUU%3D&reserved=0
> > > > ccamp-wg/draft-ietf-ccamp-layer0-types/pull/3 ; you (and other
> > > > experts) are welcome to follow up in this thread.
> > > > An updated draft have been locally created, with the diff files
> > > > attached; For detailed comments response/discussion, please see
> > > > inline starting with '[Authors]';
> > > >
> > > > Please help review if the comments are addressed, any
> > > > discussion/calls are welcome and can be arranged by chairs, thank
> you.
> > > >
> > > > Best wishes,
> > > > Haomian (on behalf of all authors & contributors)
> > > >
> > > > -----邮件原件-----
> > > > 发件人: Robert Wilton via Datatracker [mailto:noreply@xxxxxxxx]
> > > > 发送时间: 2019年12月14日 0:50
> > > > 收件人:
yang-doctors@xxxxxxxx
> > > > 抄送:
last-call@xxxxxxxx;
> > > >
draft-ietf-ccamp-layer0-types.all@xxxxxxxx;
> > > > ccamp@xxxxxxxx
> > > > 主题: Yangdoctors last call review of
> > > > draft-ietf-ccamp-layer0-types-03
> > > >
> > > > Reviewer: Robert Wilton
> > > > Review result: Almost Ready
> > > >
> > > > Comments on the document:
> > > >
> > > > 1) I would delete the "overview" paragraph at the top of section
> > > > 2, and just promote section 2.1 as section 2.
> > > > [Authors] done in -04.
> > > >
> > > > 2) 2.1. Layer 0 Types Module Contents:
> > > >
> > > > The descriptions are good, but I would suggest formatting these as
> > > > a table, or more tightly link the definition to it's description.
> > > >
> > > > E.g.
> > > >
> > > > "Operational-mode: A type that represents operational mode as
> > > > defined in [ITU-Tg6982]."
> > > >
> > > > Instead of:
> > > >
> > > > "Operational-mode:
> > > >
> > > > A type that represents operational mode as defined in [ITU-Tg6982]."
> > > > [Authors] Discussion needed: we need to be careful on changing
> > > > this, as currently the layer0-types and layer1-types are in
> > > > consistent format. Need to change in both sides or don't change any
> of them.
> > > [RW]
> > > I agree that they should be consistent between the two drafts. My
> > > aim was to make them as readable as possible, but this is purely a
> > > stylistic thing.
> > > [Authors2] Agree on only formatting that technical. Let's keep as it
> > > is, and check again if necessary.
> > >
> > > >
> > > > 3) I would define the module as YANG version "1.1" (because the
> > > > language behaviour is generally better specified) and then
> > > > reference only RFC 7950 in the introduction.
> > > > [Authors] done in -04.
> > > >
> > > > 4) typo in the Security Considerations "layer0 => layer 0", and
> > > > also in the title of section 3.
> > > > [Authors] done in -04.
> > > >
> > > > 5) I have suggested changing the module prefix to "l0-types"
> > > > rather than layer2-types. If you make this change then the IANA
> > > > considerations would need to be updated, along with section 1.2.
> > > > [Authors] done in -04.
> > > >
> > > > Comments on the YANG module:
> > > >
> > > > 1) I would suggest changing the YANG prefix to "l0-types" to help
> > > > keep it short (particularly for the identities).
> > > > [Authors] done in -04.
> > > >
> > > > 2) I would suggest making the module "yang-version 1.1;", because
> > > > the behaviour of YANG 1.1 is better specified. In fact, I would
> > > > recommend that all IETF YANG modules are defined as YANG 1.1.
> > > > [Authors] done in -04.
> > > >
> > > > 3) It is actually necessary to define frequency-thz at all? I
> > > > think that the discussion in the WG suggested that it might be
> > > > better if the frequencies are always defined in Ghz and then
> > > > converted in the client as necessary.
> > > > [Authors] We have checked all the typedef, and agreed on removing
> > > > frequency-thz and frequency-ghz together, as they are not
> > > > referenced in later groupings. Need to further check whether the
> > > > modules who import this type module would use such typedef. May
> > > > need to add back if the answer is yes. Moreover, like flexi-n,
> > > > another typedef flexi-m is added and referenced in this module.
> > > [RW]
> > > I think that defining frequency-ghz is probably still helpful, it
> > > was only frequency-thz that I was suggesting removing, effectively
> > > encouraging everyone to standardize on using frequency-ghz for layer
> > > 0 properties/considerations.
> > > [Authors2] Ok, the frequency-ghz will be brought back in upcoming -04.
> > >
> > >
> > > >
> > > > 4) frequency-ghz: Is 3 fraction digits sufficient for future
> expansion?
> > > > E.g.
> > > > It would seem to support a flex grid 3.125Ghz channel spacing, but
> > > > not
> > > > 1.5625 ...
> > > > [Authors] Discussion needed: currently the channel spacing is
> > > > 6.25GHz, and we reserve the 3.125GHz for proof-of-future.
> > > > According to Singapore discussion, WG agreed on 3.125 with 3
> fraction digits.
> > > > We prefer to keep the current format given the following
> considerations:
> > > > [Authors] Consideration 1: Technically it won't make much sense
> > > > for Layer
> > > > 0 to come down to 1.5625G, as layer 1 is providing the granularity
> > > > at ODUk
> > > > (<100Gbps) which is make very good use of such spectrums.
> > > > [Authors] Consideration 2: furthermore, if 1.5625G happens in the
> > > > future, can we keep fraction digits as 3 and specify the channel
> > > > spacing as '1.562G'? There should not be difficulties in
> > > > understanding, and we solve the problem forever.
> > > > [Authors] Consideration 3: the channel spacing is not binded with
> > > > the frequency on the grid, but a number... So maybe we can change
> > > > the dependency on the grid as well.
> > > [RW]
> > > Okay, keeping this to 3 dpi is fine.
> > >
> > > >
> > > > 5) We should aim for consistency of the identity names in the
> > > > layer-1 types module. E.g. perhaps OTU, ODU and OPU should be
> > capitalized.
> > > > [Authors] done in -04.
> > > >
> > > > 6) The models uses identities for bandwidth, but I wonder whether
> > > > defining a numerical typedef might be a better choice (e.g. more
> > > > efficient and perhaps easier for programs to work with).. Here, I
> > > > have constrained the values that are allowed, but they could also
> > > > be
> > > unconstrained:
> > > >
> > > > E.g.
> > > > typedef channel-bandwidth {
> > > > type decimal64 {
> > > > fraction-digits 2;
> > > > range
> > > > "2.66|10.70|11.04|11.09|11.27|11.31|43.01|44.57|44.58|100.00
> ...
> > > > max"
> > > > description
> > > > "Bandwidth carried by a single wavelength channel"
> > > > }
> > > > units "Gb/s";
> > > > }
> > > > Or another alternative would be to use Mb/s, which would then
> > > > allow them to be specified as a union of specific values and an
> > > > arbitrary bandwidth value.
> > > >
> > > > [Authors] Discussion needed: we hope this proposal is different
> > > > with the ones on 'identity->enumeration' but maybe push back by
> > > > extensibility, could you please confirm? Gbps is used for data
> > > > plane and should be accurate enough to be described in the YANG
> > > > module. We are open to the proposal, but need to understand the
> > > > difference and how it affects the configuration. After that we can
> > > > discuss whether decimal 64 is good, Gbps or Mbps, etc....
> > > >
> > > [RW]
> > > Identities are just names of things. So, it if the bandwidth value
> > > that clients and implementations care about, then I would expect
> > > them to need to write code to translate between the identity names
> > > and numerical values. E.g. what does a client configuring the
> > > device primarily care about here. Is it that the bandwidth is
> > > "otu2" or that it the bandwidth is "10.70G"?
> > > [Authors2] Refer to the first point raised in Aihua's feedback. On
> > > the perspective of optical network, the bandwidth for layer 0 is
> > > discrete rather than continuous. So when we specify the bandwidth,
> > > it's more a 'type configuration' than a 'value configuration', which
> > > I believe is a main difference between optical and IP. The current
> > > understanding from the authors is that the optical bandwidth is just
> > > like 'names of things' as you said, so we prefer to keep as is. BTW,
> > > for configuration, we think the proposal can reach equivalent
> > > functionality set as the current model in the draft.
> > >
> > >
> > > > 7) Same comment as for bandwidth applies to the channel spacing
> > > > identities.
> > > > I.e. I wonder whether these wouldn't be better defined using a
> > > > decimal64 type.
> > > >
> > > > E.g.
> > > > typedef dwdm-channel-spacing {
> > > > type decimal64 {
> > > > fraction-digits 2;
> > > > range
> > > > "12.5|25|50|100"
> > > > description
> > > > "Bandwidth carried by a single wavelength channel"
> > > > }
> > > > units "Ghz";
> > > > }
> > > > [Authors] Discussion needed: probably reject, see the
> > > > consideration about extensibility on layer1-types. 6.25/3.125 GHz is
> coming...
> > > [RW]
> > > The model can always be extended to support 6.25/3.125 in future.
> > > In either solution this would require a backwards compatible change
> > > to the model.
> > >
> > > My hunch is that having this as a numerical value is more useful for
> > > clients than having it as a named identity. I.e. I think that the
> > > argument for using a numerical value here is stronger than for
> > > channel bandwidth above.
> > > [Authors2] Similar as previous comments. Let me use another example
> here.
> > > In a potential GUI for the system who configure the network, it is
> > > possible to a) make a few options (100, 50, 25, 12.5, ...) in a menu
> > > format; b) leave it purely open to fill with any numbers. Given the
> > > finite values for optical configurations, the option a) would be
> > > easier with less chance to make mistakes.
> > > >
> > > > 8) Same comment as for bandwidth and dwdm-channel-spacing could
> > > > also be applied to flexi-grid-channel-spacing,
> > > > flexi-slot-width-granularity, cwdm- channel-spacing.
> > > > [Authors] Discussion needed: probably reject, see the
> > > > consideration about extensibility on layer1-types.
> > > [RW]
> > > These should be resolved with whatever the conclusion is above.
> > > [Authors2] Agree, this is our pain point now. We think all these
> > > parameters in this comment are a kind of 'type configuration' rather
> > > than 'number configuration'.
> > > >
> > > > 9) In grouping layer0-label-range-info I would rename this
> > > > grouping to l0-layer-range-info, change "layer0" => "layer 0" in the
> description.
> > > > Also "priority" could do with a more detailed description as to
> > > > what it means, etc.
> > > > [Authors] Confirmation firstly: probably a mis-spelling, propose
> > > > to
> > > > 'l0- label-range-info'.
> > > [RW]
> > > Yes.
> > > [Authors2] Okay.
> > >
> > >
> > > > [Authors] Discussion secondly: there are multiple
> > > > identities/typedef started with 'layer0-' and it should be better
> > > > to keep consistency in naming format, do we rename all of them or
> none of them?
> > > [RW]
> > > Yes, I think that I would rename all of them "l0-"
> > > [Authors2] Okay.
> > >
> > > Thanks,
> > > Rob
> > >
> > >
> > > > [Authors] priority updated in -04.
> > > >
> > > > 10) In grouping flexi-grid-label-start-end, I think that the type
> > > > should be "flexi-n" rather than "int16".
> > > > [Authors] done in -04,
> > > >
> > > > Typos: "girds" => "grids", "attrtibutes" => "attributes"
> > > > [Authors] done in -04,
> > > >
> > > > Spacing/indenting needs to be fixed:
> > > > - In "grouping wson-label-hop", just before case cwdm
> > > > - In "grouping flexi-grip-label-hop", should have a blank line
> > > > before, and "case super" block/fields indentation doesn't look
> > > > right. - In some of the typedef definitions, the "{" should move
> > > > from the start of the following line to the typedef line. In
> > > > general, as a starting point, after all other markups have been
> > > > made then it might be a good idea to use pyang to format the YANG
> > > > file for you, e.g., "pyang -f yang --yang- line-length 69", but
> > > > probably with some more blank lines, otherwise it
> > > is
> > > > a bit dense.
> > > > [Authors] done in -04, probably some compilation problems and will
> > > double
> > > > check per update.
_______________________________________________
CCAMP mailing list
CCAMP@xxxxxxxx
https://www.ietf.org/mailman/listinfo/ccamp
|
