Let me give some context:
1. GRASP is like all the good infrastructure pieces of an IGP,
aka: message passing including flooding across a domain.
Without being restricted to what IGPs where built for initially: unicast routes.
Hence the flexibility of GRASP.
2. With GRASP being a new design, it's supposedly NOT ALLOWED to simply
operate as lazily insecure as pretty much all IGPs operate today. No
security at all, or a shared password known to everybody. That's
the penalty to GRASP for being late to the game. But it would also be a good
reason to consider GRASP more than than those older designs like existing IGP
for new solutions like CATS.
3. draft-eckert-anima-acp-free-ani defines a minimum security infrastructure
for GRASP. It's a bit of a dry read, so let me explain in simpler terms:
- Security:
All nodes in a network need to trust each other,
most simply through certificates, manually installed or via BRSKI
(RFC8995) or its variations.
- Nodes run a GRASP Daemon. Think of it as an IGP daemon.
GRASP daemon discovers subnet adjacent GRASP daemons via simple
link-local multicast (DULL GRASP)
GRASP daemons build TLS connections to those neighbors
- GRAS daemons like IGP daemons forward/flood messages. In GRASP these are
freely
extensible using CBOR messages.
- For non-flooded but unicast, GRASP just uses TLS directly between the
nodes.
4. draft-eckert-anima-acp-free-ani adds also functionality where
messages need to go through firewalls or across network edges where only
different IP are supported on eithre side (e.g;: between IPv4, IPv6
domains).
I haven't yet put up this draft on the top of my list, but if there is interest
now, i am happy to work on it to support efforts that would best beenfit from
GRASP
in lightweight environments.
Cheers
Toerless
On Thu, Mar 05, 2026 at 03:31:05PM +1300, Brian E Carpenter wrote:
> Kehan,
>
> > But I think it doesn't necessarily depend on ACP. Can GRASP be implemented
> > decoupled from ACP?
>
> Technically, yes, but GRASP has no intrinsic security. For that reason, the
> GRASP standard REQUIRES a secure substrate:
> https://www.rfc-editor.org/rfc/rfc8990.html#name-required-external-security-
>
> The point is that TLS could protect most GRASP messages, but not the M_FLOOD
> and M_DISCOVERY multicasts.
>
> (In my open source implementation of GRASP I added a non-standardized
> encryption mechanism. I am not recommending this, but it proves that such
> security is possible:
> https://datatracker.ietf.org/doc/draft-carpenter-anima-quads-grasp/ . The
> code is still lurking at https://github.com/becarpenter/graspy .)
>
> Regards/Ngā mihi
> Brian Carpenter
>
> On 05-Mar-26 14:52, Yao Kehan wrote:
> > Hi Michael,
> >
> >
> > Thanks for the review.
> >
> >
> > >Yes, having CATS as an ASA seems reasonable.
> >
> > > Will you have an ACP?
> >
> >
> > [KY] The draft follows the traditional approach for deloying ASA (Establish
> > ACP channel and then signaling GRASP messages).
> >
> > But I think it doesn't necessarily depend on ACP. Can GRASP be implemented
> > decoupled from ACP? (Sorry that I have not closely followed the progress of
> > ANIMA.)
> >
> >
> > > How big are the CATS metrics?
> > > The B.2 use cases are generally thinking about objects in the hundreds of
> > > kilobytes to megabytes like .. O(2^18)
> >
> > > I would think CATS metrics would be at the O(2^10) size?
> >
> >
> > [KY] It depends on implementation. In CATS metrics definition, there are
> > three metric levels.
> > https://datatracker.ietf.org/doc/draft-ietf-cats-metric-definition/
> > <https://datatracker.ietf.org/doc/draft-ietf-cats-metric-definition/>
> >
> > The Level 0 (raw metrics) may contain many detailed compute and service
> > metrics, expressed in floating points, whose size can be large.
> >
> > For simplification and routing-friendly(easy to converge, small overhead),
> > implementers can also choose Level 1 or Level 2 normalized metrics, which
> > can be represented in 1 to 10 usigned integer value, whose size can be very
> > small.
> >
> >
> > Cheers,
> >
> > Kehan
> >
> > ----邮件原文----
> > 发件人：Michael Richardson <[email protected]>
> > 收件人：Yao Kehan <[email protected]>,anima
> > <[email protected]>,"[email protected]" <[email protected]>
> > 抄 送: (无)
> > 发送时间：2026-03-05 02:23:05
> > 主题：Re: [Anima] New draft on GRASP extension for CATS metrics distribution
> >
> >
> > Yao Kehan <[email protected]> wrote:
> > > This document describes an approach by extending GRASP signaling
> > > protocol for CATS metrics distribution.
> >
> > Yes, having CATS as an ASA seems reasonable.
> > Will you have an ACP?
> >
> > > Authors think it is aligned with the use cases like V2X described in
> > CATS use cases and requirements:
> > >
> > https://datatracker.ietf.org/doc/draft-ietf-cats-usecases-requirements/
> > > as well as
> > > https://datatracker.ietf.org/doc/draft-ietf-anima-grasp-distribution/
> >
> > How big are the CATS metrics?
> > The B.2 use cases are generally thinking about objects in the hundreds of
> > kilobytes to megabytes like .. O(2^18)
> >
> > I would think CATS metrics would be at the O(2^10) size?
> >
> > --
> > ] Never tell me the odds! | ipv6 mesh
> > networks [
> > ] Michael Richardson, Sandelman Software Works | IoT architect
> > [
> > ] [email protected] http://www.sandelman.ca/ | ruby on rails
> > [
> >
> >
> > Subject：Re: [Anima] New draft on GRASP extension for CATS metrics
> > distribution
> >
> >
> > Yao Kehan <[email protected]> wrote:
> > > This document describes an approach by extending GRASP signaling
> > > protocol for CATS metrics distribution.
> >
> > Yes, having CATS as an ASA seems reasonable.
> > Will you have an ACP?
> >
> > > Authors think it is aligned with the use cases like V2X described in
> > CATS use cases and requirements:
> > >
> > https://datatracker.ietf.org/doc/draft-ietf-cats-usecases-requirements/
> > > as well as
> > > https://datatracker.ietf.org/doc/draft-ietf-anima-grasp-distribution/
> >
> > How big are the CATS metrics?
> > The B.2 use cases are generally thinking about objects in the hundreds of
> > kilobytes to megabytes like .. O(2^18)
> >
> > I would think CATS metrics would be at the O(2^10) size?
> >
> > --
> > ] Never tell me the odds! | ipv6 mesh
> > networks [
> > ] Michael Richardson, Sandelman Software Works | IoT architect
> > [
> > ] [email protected] http://www.sandelman.ca/ | ruby on rails
> > [
> >
> >
> >
> > _______________________________________________
> > Anima mailing list -- [email protected]
> > To unsubscribe send an email to [email protected]
--
---
[email protected]
_______________________________________________
Anima mailing list -- [email protected]
To unsubscribe send an email to [email protected]
