Re: Last Call: <draft-ietf-ospf-te-metric-extensions-08.txt> (OSPF Traffic Engineering (TE) Metric Extensions) to Proposed Standard

[t.p. <[e-mail]>]

Acee

Sorry that my original comment was opaque.   Partly, I was agreeing with
Adrian that a reference to a floating point standard would be a good
idea, but disagreeing with Adrain about the particular standard, that a
more recent one was available and had been used previously by an IETF
RFC.  That part I hope is now clear, thanks to the clarifications by
others.

My tangential comment was that YANG has no facility to model floating
point numbers, having looked at doing so and rejected it, several times,
reckoning that the decimal-64, defined in RFC6020 section 9.3, is
adequate for network configuration.

A consequence of this, which is understood, is that if you think of a
number, express it in YANG's decimal-64, convert it to floating point
because that is what XPath uses (and so is used by YANG's conditional
statements), convert it back to decimal-64 then you do not get the
number you first thought of (sometimes), so a test for equality will
fail, when you might expect it to succeed.

Is this an issue?  The consensus of the netmod WG (AFAICT) is that it is
not, that the use of floating point in the IETF is minimal - after all,
it took SNMP over 20 years to get round to specifying it.

But given the explosion of interest in YANG recently, I think it is a
topic to keep an eye on so when I read an I-D and find floating point, I
point{!} out that it cannot readily be modeled.  I suspect that, in this
case, consistency with what has gone before outweighs the facility to do
it in YANG.

Tom Petch


----- Original Message -----
From: "Acee Lindem (acee)" <[e-mail]>
To: <[e-mail]>
Cc: "t.p." <[e-mail]>;
<[e-mail]>;
<[e-mail]>
Sent: Monday, December 15, 2014 4:58 PM
Subject: Re: Last Call: <draft-ietf-ospf-te-metric-extensions-08.txt>
(OSPF Traffic Engineering (TE) Metric Extensions) to Proposed Standard


Hi Adrian, Tom,
Can you guys indicate how you would like to see this comment reflected
in the draft? Are you suggesting to change he encoding to 64 bits for
the new bandwidth sub-TLVs?
Thanks,
Acee
On Dec 12, 2014, at 1:16 PM, Adrian Farrel <[e-mail]> wrote:

> Good catch Tom,
>
> Acee, the trick is to go to 6340 and look in the references :-)
>
>   [IEEE.754.2008]  Institute of Electrical and Electronics Engineers,
>                    "Standard for Floating-Point Arithmetic",
>                    IEEE Standard 754, August 2008.
>
> A
>
>> -----Original Message-----
>> From: Acee Lindem (acee) [mailto:[e-mail]]
>> Sent: 12 December 2014 18:14
>> To: t.p.
>> Cc: [e-mail];
> [e-mail];
>> [e-mail]
>> Subject: Re: Last Call: <draft-ietf-ospf-te-metric-extensions-08.txt>
(OSPF
> Traffic
>> Engineering (TE) Metric Extensions) to Proposed Standard
>>
>> Hi Tom,
>>
>> On Dec 12, 2014, at 1:08 PM, t.p. <[e-mail]> wrote:
>>
>>> On the question of Floating-Point, there is now 754-2008, which is a
>>> tighter spec and is used in RFC6340.
>>
>> What is 754-2008?
>>
>> Thanks,
>> Acee
>>
>>
>>
>>>
>>> At a tangent, the issue of floating-point support has surfaced a
number
>>> of times in YANG and, to date, has always been rejected, reckoning
that
>>> suppport for 64-bit decimal is adequate for data modelling.  The
>>> interactions with XPath (which is used as a basis for YANG
constraint
>>> statements), where floating-point is allowed, have caused a number
of
>>> discussions, some ongoing, about the comparison of a floating-point
>>> number to a 64-bit decimal one. Something to be aware of should you
ever
>>> want to model this in YANG.
>>>
>>> Tom Petch
>>>
>>>
>>> ----- Original Message -----
>>> From: "Adrian Farrel" <[e-mail]>
>>> To: <[e-mail]>
>>> Cc: <[e-mail]>; <[e-mail]>
>>> Sent: Wednesday, December 10, 2014 11:07 PM
>>>
>>>> All,
>>>>
>>>> I reviewed this document as AD and found a few small points that I
>>> have asked
>>>> the authors to address as IETF last call comments.
>>>>
>>>> Adrian
>>>>
>>>> ===
>>>>
>>>> Please look for places where you have "proposed" something and
change
>>>> that to "defined".
>>>>
>>>> ---
>>>>
>>>> It would be good to include a reference for encoding floating point
>>>> integers. The usual is (I think)...
>>>>
>>>>       IEEE, "IEEE Standard for Binary Floating-Point Arithmetic",
>>>>       Standard 754-1985, 1985 (ISBN 1-5593-7653-8).
>>>>
>>>> ---
>>>>
>>>> Section 4.2.5
>>>>
>>>>  Implementations MAY also permit the configuration of a static (non
>>>>  dynamic) offset value (in microseconds) to be added to the
measured
>>>>  delay value, to facilitate the communication of operator specific
>>>>  delay constraints.
>>>>
>>>> On the third reading I got it! I'm slow (I have a high delay :-)
>>>>
>>>> The point here is that the measured value and the static value are
>>> added
>>>> together and the sum is transmitted in this field. I'd suggest...
>>>>
>>>>  Implementations MAY also permit the configuration of a static (non
>>>>  dynamic) offset value (in microseconds) to be added to the
measured
>>>
>>>>  delay value before encoding into this TLV, to facilitate the
>>>>  communication of operator specific delay constraints.
>>>>
>>>> Similarly in 4.2.6.
>>>>
>>>> ---
>>>>
>>>> 4.2.7 appears out of sequence. But since it repeats the content of
>>>> 4.2.4 I suggest you merge them and talk about the plurality of
fields.
>>>>
>>>> ---
>>>>
>>>> Section 7
>>>>
>>>> "Sections 6 and 7 provide" should be 5 and 6.
>>>>
>>>> ---
>>>>
>>>> Section 10
>>>>
>>>>  "As per (RFC3630), unrecognized TLVs should be silently ignored"
>>>>
>>>> There has been confusion about what 3630 means by "silently
ignored".
>>>> In particular, some enthusiastic implementations thought this meant
>>> the
>>>> TLVs should be stripped from the LSA before it is propagated. I
think
>>> it
>>>> is worth the few words to explicitly state that this is not the
case.
>>>>
>>>> ---
>>>>
>>>> Section 13
>>>>
>>>> RFC 4203 is used in a normative way, please move it to the other
>>>> section.
>>>>
>>>>
>>>>
>>>
>