Subject: Re: Using a Stripped Kernel
To: Jenkins, Graham K [IBM GSA] <Graham.K.Jenkins@team.telstra.com>
From: Steven M. Bellovin <smb@research.att.com>
List: port-i386
Date: 06/05/2003 08:45:58
In message <61411576E951D211AF330008C7245DD90F9BDA88@ntmsg0005.corpmail.telstra
.com.au>, "Jenkins, Graham K [IBM GSA]" writes:
>Guys, I'm using a number of old 486's with 16Mb
>memory as Xterminals at my place, running NetBSD
>1.6.1. I'm using a cut-down kernel so as to improve
>startup and response times for X.
>
>I've found I can get a further 10% improvement by
>stripping that kernel.  But I get a couple of
>diagnostics at boot time, like:
>
>  savecore: /netbsd: kvm_nlist: bad namelist
>  savecore: /netbsd: _dumpdev:  not in namelist
>
>So .. is there a way I can compile an even smaller
>kernel so it doesn't generate these messages?
>
>These machines are operated by ordinary users, and
>they aren't the least bit interested in crash-dumps.

savecore -- and a number of other programs -- use the symbol table in
/netbsd to find certain memory locations.  Most such programs will take 
an optional argument specifying a different file -- perhaps mounted via 
NFS? -- with the symbol table.

But I'm not certain just what you're accomplishing here.  The symbol 
table affects the size of /netbsd, which is (at most) a boot-time 
effect; it won't affect run-time performance, as far as I know, because 
the symbol table doesn't occupy RAM.  What, precisely, are you 
measuring when you say "10% improvement"?  Speed?  The output of 'ls -l'?
The output of 'size'?

		--Steve Bellovin, http://www.research.att.com/~smb (me)
		http://www.wilyhacker.com (2nd edition of "Firewalls" book)