----- Original Message ----- From: "Christopher Morrow" <[email protected]> To: "Yann Berthier" <[email protected]> Cc: "sidr wg list" <[email protected]> Sent: Wednesday, March 09, 2011 3:39 AM > On Tue, Mar 8, 2011 at 8:44 PM, Yann Berthier <[email protected]> wrote: > > > > On 2011-03-08, at 4:26 PM, Christopher Morrow wrote: > > > >>> so, i suggest asking the iana for a port number, one and only one. > >>> objections? > >> > >> I'm personally a fan of keeping things simple .... what's the beef > >> with tcp/22 here? > > > > The set of source IPs needing access to tcp/22 for mgmt may not be the same as the set of IPs needing access to tcp/22 for the rpki service > > > > A dedicated port for rpki seems a good idea > > don't disagree, just seeking some justification for the change. I came up with: > o diff hosts need ssh vs rpki-rtr (potentially off-net things?) > o ability to drop shields in case of problems but still maintain > management access > o qos/debugging reasons > o authentication requirements perhaps are different between user-ssh > && rpki-rtr
I think that those are the reasons why a separate port is required. susbsystem is the standard SSH way of differentiating traffic but that is really the view of the implementer, of getting the end points of the session to the right place. Operationally, you need to be able to track the different traffic for the life of a session. RPKI is rather important, it should become critical to the Internet, so being able to offer it the best in operations seems a no brainer. As you say, filtering, performance, tracing for trouble shooting purposes, letting it through or not through firewalls etc, ... And I think it different to BGP in that it will span a wider network, whereas much of BGP is just about immediate neighbours, so there are likely to be more hops along the way where things could go wrong or need to be made right. With SNMP over SSH, there was a brief discussion about reusing the existing SNMP ports but it gained no support, rather that it was taken to be axiomatic that there would be a port, the debate being rather system or not, and one port or two (commands and notifications). Tom Petch > > randy came up with: 'maybe we don't trust the ssh code to do this > heavy lifting?' (or something similar) > > just hunting for the reason to be different. > > -chris > _______________________________________________ > sidr mailing list > [email protected] > https://www.ietf.org/mailman/listinfo/sidr _______________________________________________ sidr mailing list [email protected] https://www.ietf.org/mailman/listinfo/sidr
