Re: [Lsr] Tsvart last call review of draft-ietf-isis-te-app-13

["Les Ginsberg (ginsberg)" <[e-mail]>]

Kyle –

Inline.

From: Kyle Rose <[e-mail]>
Sent: Thursday, May 21, 2020 6:09 AM
To: Les Ginsberg (ginsberg) <[e-mail]>
Cc: [e-mail]; [e-mail]; [e-mail]; [e-mail]
Subject: Re: Tsvart last call review of draft-ietf-isis-te-app-13

On Thu, May 21, 2020 at 2:28 AM Les Ginsberg (ginsberg) <[e-mail]<mailto:[e-mail]>> wrote:
> * Update language?
>
> There are several places in which it is possible that normative language in
> prior RFCs is revised. For example, in section 6.1, the last paragraph states:
>
>   New applications which future documents define to make use of the
>   advertisements defined in this document MUST NOT make use of legacy
>   advertisements.
>
> This looks like it was written in such a way as to avoid requiring such
> updates, but it would be good to confirm that there is no normative language
> in
> older documents that would conflict with this requirement.
>
[Les:] For applications which preexisted the writing of this draft (the ones defined in Section 7.4) there are existing deployments which use legacy advertisements.
Transitioning to use new advertisements has to account for partial deployment of such support.
And the transition has to be managed by the network operator using knobs provided by implementations. This is onerous - but unavoidable in these cases.

For new applications we want to avoid this complexity/pain. So we specify new applications MUST use the new advertisements from day ONE.

As these are new applications there is no legacy to deal with and no existing specification which would be in conflict.

That's what I figured. Is it the case that new standardized applications require an RFC? If so, I think this is fine.

[Les:] Yes. In order to obtain a new bit in the new Link Attribute Application Identifier Registry defined in Section 7.4 the process defined in RFC 8126 Standards Action is required – which means an RFC has to be written for the new application.


> * Encoding
>
> In section 4.1, the bit masks are defined with a 7 bit length field for which
> only 4 bits are useful (values 0-8). It may make sense to keep the 3 high order
> bits as "reserved" and set to zero, but either way it would be nice to
> understand the justification for whatever design decision is made.
>
> You go to some length to save space (e.g., a zero-length SABM means "all
> applications") but always include an octet for UDABM length, which I
> presume
> will be zero outside the lab in most cases. How much does an extra octet
> cost?
> If it's a lot, you could use the three high-order bits to represent the first
> three applications (RSVP-TE, SRTE, and LFA) and save yourself an octet until a
> fourth application appears.
>
> For that matter, how likely are you to get to 256 standardized applications
> under IS-IS?
>
[Les:] The limitation for the xABM length to be no more than 8 bytes was introduced only recently based on a review comment.
The concern was that without a limit, it would be theoretically possible for the ABM itself to consume the full sub-TLV space, leaving no room for the actual link attribute advertisements.
So we placed a maximum length to avoid this potential issue.
As each application consumes one bit, with an 8 byte maximum length we can support 64 applications.
Sigh... yes, math is hard.

This seems more than we will ever get to - but if anyone in the WG thinks this is not large enough they should speak now so we can increase the maximum.

I suspect even 64 is safely more than you'll need, but if not there's always the possibility of a new TLV later on. :-)

There are existing implementations of this draft which have been deployed. Changing the bit positions as you suggested would not be backwards compatible.
I do not think the small space savings would be worth the trouble.

Understood, though I will point out that this illustrates the potential harm in deploying something hard to change prior to standardization. Thankfully, IS-IS is not interdomain, meaning that any such change would impose costs only on those who deployed early.
[Les:] Agreed – but it is also good to have demonstrated the viability of the extensions – as well as interoperability – before going to RFC. So early implementations have great value.
But, also, let’s not overvalue the space optimization.
IS-IS has always tried to be “frugal” in its use of space – but I think we can afford the extra byte. 😊

> In effect, use of legacy advertisements vs. app-specific advertisements is
> all-or-nothing. If that's the case, wouldn't a top-level TLV specifying a
> legacy mask for applications be more compact, less redundant, and further
> reduce the overall number/size of advertisements?
>
[Les:] The information being advertised here (link attributes) is necessarily associated with a top level TLV describing a link to a neighbor (TLVs 22, 23, 25, 141, 222, and 223) . Without that context the link attribute information is meaningless.

Yes, this makes sense, and probably would have been obvious were I more familiar with IS-IS and link state routing in general.
[Les:] Appreciate the thought you put into your review.

   Les

Thanks,
Kyle