Re: [CCAMP] Comparison of Flex-E YANG model parameters

Jiangyuanlong <jiangyuanlong@huawei.com> Thu, 30 January 2020 23:35 UTC

Return-Path: <jiangyuanlong@huawei.com>
X-Original-To: ccamp@ietfa.amsl.com
Delivered-To: ccamp@ietfa.amsl.com
Received: from localhost (localhost [127.0.0.1]) by ietfa.amsl.com (Postfix) with ESMTP id ABC2F120255; Thu, 30 Jan 2020 15:35:41 -0800 (PST)
X-Virus-Scanned: amavisd-new at amsl.com
X-Spam-Flag: NO
X-Spam-Score: -4.201
X-Spam-Level:
X-Spam-Status: No, score=-4.201 tagged_above=-999 required=5 tests=[BAYES_00=-1.9, RCVD_IN_DNSWL_MED=-2.3, SPF_PASS=-0.001] autolearn=ham autolearn_force=no
Received: from mail.ietf.org ([4.31.198.44]) by localhost (ietfa.amsl.com [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id sBs1AiSz4Cpi; Thu, 30 Jan 2020 15:35:39 -0800 (PST)
Received: from huawei.com (lhrrgout.huawei.com [185.176.76.210]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ietfa.amsl.com (Postfix) with ESMTPS id CEEDF120019; Thu, 30 Jan 2020 15:35:38 -0800 (PST)
Received: from LHREML713-CAH.china.huawei.com (unknown [172.18.7.107]) by Forcepoint Email with ESMTP id 91F0D49E98D073001BAB; Thu, 30 Jan 2020 23:35:36 +0000 (GMT)
Received: from lhreml712-chm.china.huawei.com (10.201.108.63) by LHREML713-CAH.china.huawei.com (10.201.108.36) with Microsoft SMTP Server (TLS) id 14.3.408.0; Thu, 30 Jan 2020 23:35:36 +0000
Received: from lhreml712-chm.china.huawei.com (10.201.108.63) by lhreml712-chm.china.huawei.com (10.201.108.63) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256) id 15.1.1713.5; Thu, 30 Jan 2020 23:35:36 +0000
Received: from DGGEML406-HUB.china.huawei.com (10.3.17.50) by lhreml712-chm.china.huawei.com (10.201.108.63) with Microsoft SMTP Server (version=TLS1_0, cipher=TLS_ECDHE_RSA_WITH_AES_128_CBC_SHA) id 15.1.1713.5 via Frontend Transport; Thu, 30 Jan 2020 23:35:35 +0000
Received: from DGGEML512-MBX.china.huawei.com ([169.254.2.211]) by dggeml406-hub.china.huawei.com ([10.3.17.50]) with mapi id 14.03.0439.000; Fri, 31 Jan 2020 07:35:28 +0800
From: Jiangyuanlong <jiangyuanlong@huawei.com>
To: "Rob Wilton (rwilton)" <rwilton@cisco.com>, "ccamp@ietf.org" <ccamp@ietf.org>, "draft-xiaobn-ccamp-flexe-yang-mod@ietf.org" <draft-xiaobn-ccamp-flexe-yang-mod@ietf.org>, "draft-jiang-ccamp-flexe-yang@ietf.org" <draft-jiang-ccamp-flexe-yang@ietf.org>
Thread-Topic: Comparison of Flex-E YANG model parameters
Thread-Index: AdXWxLd1rlakjne/RI6ia9o2rDqE8QA/+wDQ
Date: Thu, 30 Jan 2020 23:35:27 +0000
Message-ID: <3B0A1BED22CAD649A1B3E97BE5DDD68BD389B62D@dggeml512-mbx.china.huawei.com>
References: <MN2PR11MB43669F3E38B55BFDB457D768B5050@MN2PR11MB4366.namprd11.prod.outlook.com>
In-Reply-To: <MN2PR11MB43669F3E38B55BFDB457D768B5050@MN2PR11MB4366.namprd11.prod.outlook.com>
Accept-Language: zh-CN, en-US
Content-Language: zh-CN
X-MS-Has-Attach:
X-MS-TNEF-Correlator:
x-originating-ip: [10.210.167.145]
Content-Type: text/plain; charset="Windows-1252"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0
X-CFilter-Loop: Reflected
Archived-At: <https://mailarchive.ietf.org/arch/msg/ccamp/4XgKJw2EjVmbQC-fontJA74Jq18>
Subject: Re: [CCAMP] Comparison of Flex-E YANG model parameters
X-BeenThere: ccamp@ietf.org
X-Mailman-Version: 2.1.29
Precedence: list
List-Id: Discussion list for the CCAMP working group <ccamp.ietf.org>
List-Unsubscribe: <https://www.ietf.org/mailman/options/ccamp>, <mailto:ccamp-request@ietf.org?subject=unsubscribe>
List-Archive: <https://mailarchive.ietf.org/arch/browse/ccamp/>
List-Post: <mailto:ccamp@ietf.org>
List-Help: <mailto:ccamp-request@ietf.org?subject=help>
List-Subscribe: <https://www.ietf.org/mailman/listinfo/ccamp>, <mailto:ccamp-request@ietf.org?subject=subscribe>
X-List-Received-Date: Thu, 30 Jan 2020 23:35:42 -0000

Hi Rob,

Thanks a lot for the detailed comments, I agree with most of them (except several minor items, I will comment on them later).
I will be glad to work with you and the other authors on a common FlexE model draft.

Cheers,
Yuanlong

-----Original Message-----
From: Rob Wilton (rwilton) [mailto:rwilton@cisco.com] 
Sent: Thursday, January 30, 2020 12:58 AM
To: ccamp@ietf.org; draft-xiaobn-ccamp-flexe-yang-mod@ietf.org; draft-jiang-ccamp-flexe-yang@ietf.org
Subject: Comparison of Flex-E YANG model parameters

Hi,

Sorry for the delay.

After I think that we reached rough agreement on Flex-E model structure, I have now (with my limited knowledge) gone through the flex-e configuration parameters in draft-xiaobn and draft-jiang.  I think that I have hopefully given comments on all areas of the configuration.  I think that some aspects will be easier to resolve and reach conclusion on than other parts.

Hence, to try and make forward progress, I suggest that we try and defer getting stuck in discussion on the following topics:

  1.  Whether we need to have explicit configuration for both A and B calendars or not.  The model will definitely be simpler for client if this is not required, so I would recommend that we assume that it is not required, try and resolve the other points and then come back to this potentially thorny issue.
  2.  Whether we need to have the "expected" configuration leaves.  I'm not sure I fully understand the intent of this, but I suspect that the intent is to use the calendar protocol as a way of checking that both sides are correctly configured.  My suggestion, to simplify the model, is to assume that in this case that the RX and TX calendar for client interfaces are the same, and hence this configuration could probably be expressed through a different "calendar-protocol-enable/tx-calendar-neg" setting.  But again, I suspect that this might be another can or worms that we might want to try and keep the lid on for the moment whilst we get agreement for the other parts of the model.  But, of course, this could be handled through a vendor or standard augmentations, or YANG features, but I still have a preference is trying to keep the base model as simple as we can.

Barring these two points, you will see that I have tried to give a suggestion for the other configuration items, with an aim to try and get a compromise starting model that the authors and WG can move forward with.

Based on the comments that I receive, I propose updating the YANG model here (https://github.com/rgwilton/flex-e-yang).  If we do manage to find common ground over the next week or two, hopefully it will be possible for both sets of authors to take the combined model, polish it, and work together on a common FlexE model draft?

FlexE Group, Bonded-phy, instance and client configuration comparison:

(A) Group level configurable attributes:
    (A 1) Both drafts define a per group slot granularity (slot-granularity/cal-slot-gran)
     - both represented by an enum
     - both allow 5g and 25g (as per the OIF standard)
     - draft-jiang draft also allows "slot-others".
     - draft-jiang defines a default value (5g), but draft-xiaobn does not.
     - My recomendation:
       - The base model should only include 5g and 25g, because that is what is in the spec, and vendors can always deviate this leaf if required to support non-standard values.
       - I think that a default value is useful and should be defined, but possibly using 25g as the default because it is more forward looking (as bandwidths increase).

    (A 2) Both drafts support configuring Flex-E phy type (100g, 200g, 400g, 50g) (flexe-phy-type)
     - I'm not sure that this configuration is required, or just useful as an extra check.
     - Vendors could always deviate add "mandatory" if required.
     - I suggest that this leaf should be optional, and have no default value defined.

    (A 3) draft-xiaobn supports limiting the bandwidth for a flex-e group (flexe-gp-avb-bw)
     - This configuration seems to act as an additional constraint.  I don't see any harm in this configuration, but it seems somewhat redundant (given the bandwidth can be calculated for each member interface), and hence as a starting point, my preference would be to leave this out of a standard FlexE configuration data model.

    (A 4) Both drafts seem to have a mechanism to control the calendar protocol (calendar-protocol-enable/tx-calendar-neg):
      - I would suggest changing the name of this leaf to "calendar-mode", and use an enumeration for future extensibility.
      - Enumeration choices could be
         "Static", - calendar configuration is static (and assumed to be the same for both RX and TX), calendar protocol information is ignored.
         "Verified" - calendar configuration is static (and assumed to be the same for both RX and TX), calendar protocol information is checked, otherwise the client interface is kept down in an errored state?  E.g. replacing the need for the "expected" configuration values in draft-xiaobn.
         "Master-Slave" calendar configuration is static for TX, RX is determined from the calendar protocol (although perhaps should use different terminology for political correctness)
      - Having a default value for the calendar protocol makes sense, but I don't know what the appropriate default is.

    (A 5) draft-xiaobn defines reply-ca-mode:
      - I'm not quite sure what this leaf is for, and hence I question whether it is required since it appears to be somewhat implementation specific.
(B) Bonded-phy level configuration:
    (B 1) draft-xiaobn contains a "remote-phy-interface" as an interface-ref.  This configuration cannot work as an interface-ref (because it would attempt to validate against the local device's interface list).  I suspect that this isn't really configuration at all, but possibly just documentation.  As a starting point, my preference would be to leave this out of a standard FlexE configuration data model.  Vendors could always augment this in, with little interop issues.
    (B 2) Both drafts define mechanisms under the bonded-phy/[instance for draft-xiaobn] to specify the unavailable slots.  Rather than defining these as lists of unavailable slots, I would merge these into a single per bonded-phy leaf of ranges of what are the available slots, and if the leaf is not configured then all slots (for available instances) can be used.
        E.g. I am suggesting a per bonded-phy leaf:
            available-calender-slots = "1-15, 19, 20-29"
(C) Instance configuration:
    (C 1) draft-xiaobn has explicit support for flex-e "instances" from FlexE 2.1,(e.g. a 200G PHY is represented as 2x100G instances, a 400G PHY, has 4x100 instances).
     - I would think that any instance configuration should naturally be under the PHY, since from the spec, it seems to be phy specific, and relative to the phy.
     - FlexE has the ability to disable instances, but it must always be the highest instance numbers on a phy that are disabled.
        * One option is that this could be represented by a single configurable uint8 under the bonded-phy that indicates the number of supported-instances and defaults to all instances being supported if not configured.  [Note - this configuration is probably equivalent to the "uneq-flexe-instance", but in reverse sense and represented in a more concise way.]
        * An alternative option would be to infer which instances are disabled from the available-calendar-slot list.  E.g. for a 400G interfaces, stating available-slots="1-39" would imply that only instances 1 and 2 are in use, and instances 3 and 4 are disabled.
     - Otherwise I don't think that it is necessary to configure/define a separate calendar slot availability per instance.  I think that it will be more concise and simpler for clients if this is just configured under the bonded phy.  E.g., assuming 5G slots,  instance 0 would be slots 0-19, instance 1 would be slots 20-39, instance 2 would be slots 40-59, etc...
(D) "Expected configuration":
  (D 1) draft-xiaobn defines expected-group-number, expected-phy-map, expected-cal-cfg.
    - I don't think that these configuration items should be required.  These settings should either just be (i) regular configuration, i.e. covered by the other configuration items, or learned from the peer (e.g. if it can be carried via protocol).

(E) "Calendar configuration"
  (E 1) draft-xiaobn defines tx-calendar, rx-calendar (and also tx-calendar-neg but that has been covered above).
   - I think that it is useful to be able to configure whether the A or B calendar is being used, defaulting to "A" if not specified.
   - I'm not convinced that the standard model should have rx-calendar. In the case that it is statically configured, I think that the standard model should assume that the RX and TX calendar are the same.


(F) Client interface configuration
(F 1) From reading the OIF-FlexE spec, on the wire it looks like clients are bound to instances, but I don't think that this should be in the management model, it feels like it would be exposing unnecessary complexity.
   - draft-jiang allows each individual slot to be assigned to assigned to a phy and a slot on that phy.  This could end up being verbose.  E.g. 20 entries for a 100G client interface with 5G slot sizes.  Client ordering is explicitly specified by the client-slot-id leaf.
   - draft-xioabn maps to instance number and slot-id.  Client ordering isn't specified.  This would probably work if the list is "ordered-by user".
   - In both cases, I think that the configuration is a bit verbose, particularly for simple/normal configurations.  Hence, I would recommend a more compact format:
     - An ordered-by user list of phys, where for each phy the slots are specified in a range string, e.g. "1-10, 15-19".  The mapping from the client slot to phy slot is implicit from the client ordering of the phys and the slots.
     - This does effectively impose some restrictions (or simplifications):
       - The slots used by a client on a given phy are always in order, e.g. I think that "10-1" or "6,3,2,9,5" should not be allowed.
       - You can't have some slots on one phy, then some on another phy, and then more on the first phy again.  All the slots on a given phy must be specified together (because the phy interface would be the list key).

(F 2) Explicit configuration for calendar A vs calendar B.
  - draft-xioabn allows both A and B calendars to be specified separately, draft-jiang does not.
  - My view is that configuring both A and B calendars complicates the model, and I would prefer that it is left out of an initial joint WG draft, but I understand that significant further discussion on this point is probably required.

(F 3) rx calendar slots and expected calendar slots.
  - draft-xioabn allows, if the tx-calendar is in static mode, to specify the rx-calendar slots, and the expected A and B calendar slots.
  - I think that in static mode we should probably assume that RX calendar slots are the same as the TX calendar slots.
  - I don't think that the expected slots configuration should be required, or perhaps could be better modelled via the calendar protocol mode.

(F 4) Bandwidth
   - draft-xiaobn defines client inteface bandwdth both in terms of the signal rate and mac-rate.
   - I think that having client interface bandwidth is required, and probably appropriate to define here under flexe client interface configuration.
   - Specifying the bandwidth in Gbps is probably sufficient, the signal rate can be inferred from the bandwidth.  Allowable values (10, 40, or m*25).


(G) Operational values:
(G 1) draft-jiang defines flexe-phy-status:
   - I think that this is okay, although I question whether this would end up being a duplicate of information already available in a generic Ethernet Phy YANG model.
   - I propose that this is retained at this time.

(G 2) draft-jiang defines flexe-client-status:
   - I think that this is a duplicate of the status information that should be available under the clinet interface itself for all Ethernet interfaces