Hi Jens,

Thanks for the comments, please see inline.

Best,

Dirk

-----Original Message-----
From: rtgwg <[email protected]> On Behalf Of Jens Finkhaeuser
Sent: 27 October 2022 20:37
To: [email protected]
Subject: Re: FW: New Version Notification for draft-trossen-rtgwg-rosa-00.txt

Hi all,

I have a few smaller comments, I think.

1. Since I've been pulled into icnrg, I can't help but notice that ROSA could 
work for ICN. The only concern here is related to naming; in ROSA, "services" 
are addressed, in ICN it's "named data objects". Maybe a more generic term like 
"resource" is better? At any rate, reference to ICN could be made.

(ROSA = Routing on Service Addresses)
(ROAR = Routing on Addresses [for] Resources)

[DOT] I like ROAR 😉 Indeed, ROSA has a provenance in concepts of ICN (which may 
not be surprising considering previous work of mine) and a longer document on 
ROSA includes this more clearly. In this draft, the clearest linkage comes 
through the proposed usage of RFC8609 for constructing the service address. The 
approach in ROSA though is to look at the problem overall as one of service 
delivery, so even if you move from, say URL type names to resource names, the 
view on 'services' remains (as in invocating some sort of transactional 
interaction with the destination, which may be the choice of one of possibly 
many). So even if you were to use ROSA with rather fine-grained, say 
chunk-level naming, the service would be that of 'retrieving the chunk' and the 
problem remains that of selecting one of possibly many locations where that 
chunk may reside. Again, this is similar to what the ICN community has been 
working on but ROSA tries to embody this in a service-centric approach over an 
IPv6 EH shim overlay to make the routing itself tractable (since routing 
entries in the intermediary SARs are limited to those 'services' that have a 
relationship with the ROSA provider and do not include generally ALL services 
possible to invoke). 

2. Pinning to a service instance/locator as described in 5.4 is of course 
reasonable for TCP, etc. but is slightly at odds with multi-homing. If the 
service is multi-homed, it would make sense to announce all of its locators as 
belonging to the same instance. In-path routers can then proceed as before. But 
if the client is multi-homed as well, it may need to be/run its own ingress SAR 
to decide on the best link. For the smoothest failover, it should be able to 
make packet-by-packet routing decisions without disrupting the session.

All of which is to say that pinning to an instance and pinning to a locator 
might best be treated separately. That is (going back to 5.2):

- announce multiple instance IPs (multi-homed service) along with an instance ID
- in service responses, include the the instance ID and currently selected 
instance IP
- then in the socket layer, pin to the instance ID and let a local SAR resolve 
that to the instance IP based on local multi-homing knowledge

[DOT] I fully agree with your reasoning and we had similar thoughts, including 
also the support for multicast (we have some parallel work, also presented 
partially to the IETF already, on DLTs, assessing the impact of DLTs on 
networks with the intention of using ROSA concepts to realize the costly 
diffusion multipoint operation so inherent for many DLT platforms). Coming back 
to your suggestion, this is absolutely possible. My worry though was that 
complicating the operations of the SAR (as you say '...to decide on the best 
link') was not the best starting point for me in an initial draft. So we left 
multi-homing aspect out of this initial version. But I am happy to revisit this 
(and will come back to you explicitly on this) if this work will progress 
positively. 

This, however, can also be added in a later extension.

3. On 5.6 (Interconnection), there are other options than using DNS. In either 
case, that only works for DNS-compatible service identifiers, which may be what 
you intend with the new socket (address) type you describe. But it would for 
e.g. use in ICN impose unnecessary constraints. Maybe making that resolution 
step an implementation detail of the SAG. A simple scheme prefix for service 
identifiers could be used to distinguish the preferred resolution methods, 
while keeping the option of using non-DNS-compatible service identifiers.

[DOT] indeed, I should have made it clearer that our focus is the 
interconnection to DNS-exposed services and via the DNS. You could also 
envision the use of private mapping services, in which case ROSA will become a  
limited domain technology not interconnecting with the Internet at all, yet 
allowing for connecting to other ROSA domains. 

None of which speaks against the proposal, really.
[DOT] Thanks for the positive view on this work!

Jens

> -----Original Message-----
> From: rtgwg [email protected] On Behalf Of Dirk Trossen
> Sent: 24 October 2022 11:02
> To: [email protected]
> Subject: FW: New Version Notification for 
> draft-trossen-rtgwg-rosa-00.txt
> 

> Dear all,
> 

> We (Luis from Telefonica and myself) have posted an I-D on 'routing on 
> service addresses' (see below), outlining an approach to direct IP packets 
> based on service address information rather than network locator addresses in 
> an end-to-end manner. For this, we suggest the use of IPv6 extension headers, 
> utilized by a shim overlay atop IPv6 and realized by what we call a ROSA 
> provider.
> 

> We have requested a presentation slot for the upcoming IETF meeting in London 
> but I would love to hear feedback and comments already on the list, if 
> possible.
> 

> I am also looking forward to any discussions at the IETF, including as side 
> discussions, with those interested in this topic.
> 

> Many thanks!
> 

> Best,
> 

> Dirk
> 

> -----Original Message-----
> > From: [email protected] [email protected]
> > 

> > Sent: 24 October 2022 10:56
> > To: Luis M. Contreras [email protected]; 
> > Dirk Trossen [email protected]; Luis Contreras 
> > [email protected]
> > 

> > Subject: New Version Notification for 
> > draft-trossen-rtgwg-rosa-00.txt
> > 

> > A new version of I-D, draft-trossen-rtgwg-rosa-00.txt has been successfully 
> > submitted by Dirk Trossen and posted to the IETF repository.
> > 

> > Name: draft-trossen-rtgwg-rosa
> > Revision: 00
> > Title: Routing on Service Addresses
> > Document date: 2022-10-24
> > Group: Individual Submission
> > Pages: 31
> > URL: https://www.ietf.org/archive/id/draft-trossen-rtgwg-rosa-00.txt
> > Status: https://datatracker.ietf.org/doc/draft-trossen-rtgwg-rosa/
> > Htmlized: 
> > https://datatracker.ietf.org/doc/html/draft-trossen-rtgwg-rosa
> > 

> > Abstract:
> > This document proposes a novel communication approach which reasons 
> > about WHAT is being communicated (and invoked) instead of WHO is 
> > communicating. Such approach is meant to transition away from 
> > locator-based addressing (and thus routing and forwarding) to an 
> > addressing scheme where the address semantics relate to services 
> > being invoked (e.g., for computational processes, and their 
> > generated information requests and responses).
> > 

> > The document introduces Routing on Service Addresses (ROSA), as a 
> > realization of what is referred to as 'service-based routing' (SBR).
> > Such routing is designed to be constrained by service-specific 
> > parameters that go beyond load and latency, as in today's best 
> > effort or traffic engineering based routing, leading to an approach 
> > to steer traffic in a service-specific constraint-based manner.
> > 

> > Particularly, this document outlines sample ROSA use case scenarios, 
> > requirements for its design, and the ROSA system design itself.
> > 

> > The IETF Secretariat
> > 

> > _______________________________________________
> > rtgwg mailing list
> > [email protected]
> > https://www.ietf.org/mailman/listinfo/rtgwg
_______________________________________________
rtgwg mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/rtgwg

Reply via email to