On Fri, Feb 05, 1999 at 01:20:13PM +0100, Anders Magnusson wrote:
> > 
> > > Sounds OK to me though I also like Paul Goyette's idea:
> > > 
> > > 	sysctl -w net.inet.ip.<ifname>.srcaddr=NNN.NNN.NNN.NNN
> > 
> > Unfortunatly you can't do that w/ sysctl. All the parts of the sysctl MIB
> > need to be known at compile time - of sysctl!
> > 
> Actually, the sysctl interface should/will be redesigned. Loading of
> modules that can have sysctl-modifying functionality isn't possible
> today because of this limitation.

Good!  When this is done, could some care be taken to ensure that the
current proliferation of fooctl programs can be eliminated?

If we're going to present a MIB, we should present *one* damned MIB,
not one for "sys", one for "audio", one for "ata", and so forth.  Or
stated a bit differently, I'm not so sure that having even _one_
namespace outside the filesystem namespace (the first such offender was
the socket interface) was a good idea -- but I'm VERY sure that having
500 such namespaces is a bad idea.  If we can't *be* plan 9 in this
regard, at least we could try to be a bit more like it.

The point of sysctl as I understand it is to support SNMP-like management
of the machine.  Having N 'ctl' utilities with separate MIBs completely
undermines this -- those separate MIBs should live under the sysctl MIB
somewhere.

-- 
Thor Lancelot Simon	                                      tls@rek.tjls.com
	"And where do all these highways go, now that we are free?"