sowmini.varadhan at sun.com wrote:
> However, the idea was to have an administrative interface that was
> approximately one per OSI layer, and the motivation was that these
> interfaces, at least for the TCP/IP family of protocols, would
> operate on a similar set of objects.
>
> I'd like to hear other thoughts around this. If we have tcpadm and udpadm,
> then what about sctpadm, dhcpadm and all the other protocols? Where
> do we stop the explosion?
At which layer does DHCP, router discovery RIP, BGP, etc operate?
They are protocols which use some communication service (UDP, ICMP, and
TCP) to carry packets around i.e. they operate at >layer 4, but they
provide configuration which is used by layer 3 hence they are <layer 3.
The OSI model is just that - a model - with its simplifications and
limitations.
It doesn't describe the modularity and dependencies of a system.
For IP we have a set of objects that can be created, destroyed, with
properties (IP addresses, etc).
But for the transports all we have is a set of properties.
The latter can be handled with a "sysctl-like" approach where the name
of the property is hiearchical e.g. tcp/tcp_wait_timer.
For components other that IP which have objects which can be created and
destroyed it might make sense to introduce a separate command to keep
things clean; hard to predict. Do we have examples of such components?
Erik
