Re: A little rant over draft-fenner-zinin-rtg-standard-reqts-01

classic Classic list List threaded Threaded
8 messages Options
Reply | Threaded
Open this post in threaded view
|

Re: A little rant over draft-fenner-zinin-rtg-standard-reqts-01

Bill Fenner

>This now frees us up to examine the requirement for *two* independent
>implementations.

While you have some excellent points, our reasoning for changing the
requirement to 2 implementations was nowhere near that interesting.
The existing requirements for PS in 1264 were:

a) At least one implementation.
b) The spec must be written clearly enough to permit two interoperable
implementations to be written from it.

After trying to apply these guidelines for a while, we realized
that we had no real way to determine "b", and decided to switch
to 2 implementations to prove the clarity of the spec.

>From my perspective this debate reflects a change to the way that a PS is
>cast within the industry. I think that at the moment it is hard to build
>significant operational experience until publication as PS.

I think this experience varies - there are a fair number of things
that are implemented and deployed before they're I-Ds.

>A solution to this, is to introduce a new category between WG last call
>and PS. Let's call it PPS (proposed proposed standard). We would then
>advance I-Ds to PPS, allow implementation and operational experience
>before advancing to PS.

This sounds like what has been tossed around in newtrk as
"stable snapshot".

>Perhaps we should try to understand these concerns (personally, I think
>the Motivation section is fine, but I don't see the text there being
>mapped to concrete concerns, and I don't see a description of how the
>proposed solution will address the text in the Motivation section).
>1. "The Internet is being built from PSs"
>2. "There is insufficient operational experience of protocols that are
>being deployed in the Internet"

We're explicitly not trying to address #1.

It's not that we're trying to fix #2, but prevent it.

We[*] had the conversation some time ago about whether it's still
reasonable for routing protocols to have their own rules.  The
general feeling seemed to be that they do, mostly for the reasons
outlined in the motivation section.  Personally, I'm not a huge
fan, and would rather see IETF-wide consistency (e.g., if DNS
goes kaplooey, is it a bigger or lesser catastrophe for the Internet
than if routing goes kaplooey?), but the general feeling of the
community seemed to be until we have that, at least we have this.
The effort to revise 1264 came as an attempt to modernize it.

  Bill

[*] - at a routing area meeting a few years ago.  I don't think there
was followup on the list at the time.

_______________________________________________
routing-discussion mailing list
[hidden email]
https://www1.ietf.org/mailman/listinfo/routing-discussion
Reply | Threaded
Open this post in threaded view
|

Re: A little rant over draft-fenner-zinin-rtg-standard-reqts-01

Alex Zinin

Pushing this forward a little...

My reading of rough consensus regarding this discussion (this thread and
"comment on draft-fenner-zinin-rtg-standard-reqts-00" started on
Aug-3-2005) is that we should require two implementations for PS.

--
Alex
http://www.psg.com/~zinin

Monday, November 21, 2005, 3:07:08 PM, Bill Fenner wrote:

>>This now frees us up to examine the requirement for *two* independent
>>implementations.

> While you have some excellent points, our reasoning for changing the
> requirement to 2 implementations was nowhere near that interesting.
> The existing requirements for PS in 1264 were:

> a) At least one implementation.
> b) The spec must be written clearly enough to permit two interoperable
> implementations to be written from it.

> After trying to apply these guidelines for a while, we realized
> that we had no real way to determine "b", and decided to switch
> to 2 implementations to prove the clarity of the spec.

>>From my perspective this debate reflects a change to the way that a PS is
>>cast within the industry. I think that at the moment it is hard to build
>>significant operational experience until publication as PS.

> I think this experience varies - there are a fair number of things
> that are implemented and deployed before they're I-Ds.

>>A solution to this, is to introduce a new category between WG last call
>>and PS. Let's call it PPS (proposed proposed standard). We would then
>>advance I-Ds to PPS, allow implementation and operational experience
>>before advancing to PS.

> This sounds like what has been tossed around in newtrk as
> "stable snapshot".

>>Perhaps we should try to understand these concerns (personally, I think
>>the Motivation section is fine, but I don't see the text there being
>>mapped to concrete concerns, and I don't see a description of how the
>>proposed solution will address the text in the Motivation section).
>>1. "The Internet is being built from PSs"
>>2. "There is insufficient operational experience of protocols that are
>>being deployed in the Internet"

> We're explicitly not trying to address #1.

> It's not that we're trying to fix #2, but prevent it.

> We[*] had the conversation some time ago about whether it's still
> reasonable for routing protocols to have their own rules.  The
> general feeling seemed to be that they do, mostly for the reasons
> outlined in the motivation section.  Personally, I'm not a huge
> fan, and would rather see IETF-wide consistency (e.g., if DNS
> goes kaplooey, is it a bigger or lesser catastrophe for the Internet
> than if routing goes kaplooey?), but the general feeling of the
> community seemed to be until we have that, at least we have this.
> The effort to revise 1264 came as an attempt to modernize it.

>   Bill

> [*] - at a routing area meeting a few years ago.  I don't think there
> was followup on the list at the time.

> _______________________________________________
> routing-discussion mailing list
> [hidden email]
> https://www1.ietf.org/mailman/listinfo/routing-discussion


_______________________________________________
routing-discussion mailing list
[hidden email]
https://www1.ietf.org/mailman/listinfo/routing-discussion
Reply | Threaded
Open this post in threaded view
|

draft-fenner-zinin-rtg-standard-reqts-01: operational experience for PS?

Alex Zinin
A related question that I saw in the discussion on the number of
implementations for PS was wrt the operational experience.

RFC 1264, as well as the draft-fenner-zinin say no experience is necessary
for PS. There was a suggestion by JP, seconded by Curtis that at least
minimal operational experience should be required. I'd like to hear
opinions on this. Please keep in mind, that requiring this also means
more paperwork for submission, however small incrementally.

--
Alex
http://www.psg.com/~zinin

Tuesday, February 14, 2006, 5:25:48 PM, Alex Zinin wrote:

> Pushing this forward a little...

> My reading of rough consensus regarding this discussion (this thread and
> "comment on draft-fenner-zinin-rtg-standard-reqts-00" started on
> Aug-3-2005) is that we should require two implementations for PS.



_______________________________________________
routing-discussion mailing list
[hidden email]
https://www1.ietf.org/mailman/listinfo/routing-discussion
Reply | Threaded
Open this post in threaded view
|

Re: A little rant over draft-fenner-zinin-rtg-standard-reqts-01

Curtis Villamizar
In reply to this post by Alex Zinin

In message <[hidden email]>
Alex Zinin writes:

>  
>  
> Pushing this forward a little...
>  
> My reading of rough consensus regarding this discussion (this thread and
> "comment on draft-fenner-zinin-rtg-standard-reqts-00" started on
> Aug-3-2005) is that we should require two implementations for PS.
>  
> --
> Alex


Two *independent* implementations.

Curtis

_______________________________________________
routing-discussion mailing list
[hidden email]
https://www1.ietf.org/mailman/listinfo/routing-discussion
Reply | Threaded
Open this post in threaded view
|

Re: draft-fenner-zinin-rtg-standard-reqts-01: operational experience for PS?

Joel M. Halpern
In reply to this post by Alex Zinin
I think that it is reasonable to ask folks to implement pre-standard
for routing activities. (If we can't get two implementations, then
there probably isn't enough interest to be worth standardizing.)

But asking for operational deployment pre-standard is a lot
harder.  Yes, some of our work has gotten deployment before PS.  But
not all of it by any means.

The paperwork increment is probably not an issue.  The real world
requirement for use (rather than just implementation) seems to me a
mistake.  I might even feel tempted to go a step further.  Even we
put in this requirement, then we are essentially declaring the last
WG draft to be the same as what everyone else treats as PS, without
IETF-wide review.

Yours,
Joel M. Halpern

At 08:55 PM 2/14/2006, Alex Zinin wrote:

>A related question that I saw in the discussion on the number of
>implementations for PS was wrt the operational experience.
>
>RFC 1264, as well as the draft-fenner-zinin say no experience is necessary
>for PS. There was a suggestion by JP, seconded by Curtis that at least
>minimal operational experience should be required. I'd like to hear
>opinions on this. Please keep in mind, that requiring this also means
>more paperwork for submission, however small incrementally.
>
>--
>Alex
>http://www.psg.com/~zinin
>
>Tuesday, February 14, 2006, 5:25:48 PM, Alex Zinin wrote:
>
> > Pushing this forward a little...
>
> > My reading of rough consensus regarding this discussion (this thread and
> > "comment on draft-fenner-zinin-rtg-standard-reqts-00" started on
> > Aug-3-2005) is that we should require two implementations for PS.
>
>
>
>_______________________________________________
>routing-discussion mailing list
>[hidden email]
>https://www1.ietf.org/mailman/listinfo/routing-discussion


_______________________________________________
routing-discussion mailing list
[hidden email]
https://www1.ietf.org/mailman/listinfo/routing-discussion
Reply | Threaded
Open this post in threaded view
|

Re: draft-fenner-zinin-rtg-standard-reqts-01: operational experience for PS?

tony.li


> I think that it is reasonable to ask folks to implement pre-standard for
> routing activities. (If we can't get two implementations, then there
> probably isn't enough interest to be worth standardizing.)
>
> But asking for operational deployment pre-standard is a lot harder.
> Yes, some of our work has gotten deployment before PS.  But not all of
> it by any means.
>
> The paperwork increment is probably not an issue.  The real world
> requirement for use (rather than just implementation) seems to me a
> mistake.  I might even feel tempted to go a step further.  Even we put
> in this requirement, then we are essentially declaring the last WG draft
> to be the same as what everyone else treats as PS, without IETF-wide
> review.


Slightly different perspective: how would one truly define meaningful
operational experience?  I don't think that any of us would accept 10
minutes in a vendor's back lab.  To my way of thinking, the meaningful
experience only comes with many thousands of hours of experience in
production networks, facing real world churn, topology changes, corner
cases, and scalability stresses.  How one would articulate that
precisely enough seems somewhat challenging, and how one would actually
ascertain that level of experience also seems hard.

Thus, rather than add a well-intentioned but ineffective requirement,
I'd suggest that we pass on it.

Tony

_______________________________________________
routing-discussion mailing list
[hidden email]
https://www1.ietf.org/mailman/listinfo/routing-discussion
Reply | Threaded
Open this post in threaded view
|

Re: draft-fenner-zinin-rtg-standard-reqts-01: operational experience for PS?

Chandrashekhar Appanna
In reply to this post by Alex Zinin
  Where would we capture the operational experience? In another
  document like the implementation report or in the same one?
  I like the idea but I am not clear what we would actually
  quantify it.. otherwise it just becomes another 'grey' area.

  Rgds,
  Chandra.
 
On Tue, Feb 14, 2006 at 05:55:05PM -0800, Alex Zinin wrote:

> A related question that I saw in the discussion on the number of
> implementations for PS was wrt the operational experience.
>
> RFC 1264, as well as the draft-fenner-zinin say no experience is necessary
> for PS. There was a suggestion by JP, seconded by Curtis that at least
> minimal operational experience should be required. I'd like to hear
> opinions on this. Please keep in mind, that requiring this also means
> more paperwork for submission, however small incrementally.
>
> --
> Alex
> http://www.psg.com/~zinin
>
> Tuesday, February 14, 2006, 5:25:48 PM, Alex Zinin wrote:
>
> > Pushing this forward a little...
>
> > My reading of rough consensus regarding this discussion (this thread and
> > "comment on draft-fenner-zinin-rtg-standard-reqts-00" started on
> > Aug-3-2005) is that we should require two implementations for PS.
>
>
>
> _______________________________________________
> routing-discussion mailing list
> [hidden email]
> https://www1.ietf.org/mailman/listinfo/routing-discussion

_______________________________________________
routing-discussion mailing list
[hidden email]
https://www1.ietf.org/mailman/listinfo/routing-discussion
Reply | Threaded
Open this post in threaded view
|

Re: draft-fenner-zinin-rtg-standard-reqts-01: operational experience for PS?

Alex Zinin
Chandra-

 At this point we put it in an "experience with the protocol" document. for
 DS and higher.

--
Alex
http://www.psg.com/~zinin

Wednesday, February 15, 2006, 6:00:01 AM, Chandrashekhar Appanna wrote:
>   Where would we capture the operational experience? In another
>   document like the implementation report or in the same one?
>   I like the idea but I am not clear what we would actually
>   quantify it.. otherwise it just becomes another 'grey' area.

>   Rgds,
>   Chandra.
 

> On Tue, Feb 14, 2006 at 05:55:05PM -0800, Alex Zinin wrote:
>> A related question that I saw in the discussion on the number of
>> implementations for PS was wrt the operational experience.
>>
>> RFC 1264, as well as the draft-fenner-zinin say no experience is necessary
>> for PS. There was a suggestion by JP, seconded by Curtis that at least
>> minimal operational experience should be required. I'd like to hear
>> opinions on this. Please keep in mind, that requiring this also means
>> more paperwork for submission, however small incrementally.
>>
>> --
>> Alex
>> http://www.psg.com/~zinin
>>
>> Tuesday, February 14, 2006, 5:25:48 PM, Alex Zinin wrote:
>>
>> > Pushing this forward a little...
>>
>> > My reading of rough consensus regarding this discussion (this thread and
>> > "comment on draft-fenner-zinin-rtg-standard-reqts-00" started on
>> > Aug-3-2005) is that we should require two implementations for PS.
>>
>>
>>
>> _______________________________________________
>> routing-discussion mailing list
>> [hidden email]
>> https://www1.ietf.org/mailman/listinfo/routing-discussion


_______________________________________________
routing-discussion mailing list
[hidden email]
https://www1.ietf.org/mailman/listinfo/routing-discussion