Subject: Re: vmstat, iostat etc no longer work?
To: Bill Studenmund <skippy@macro.stanford.edu>
From: Grey Wolf <greywolf@siva.captech.com>
List: current-users
Date: 11/13/1996 11:50:03
Bill Studenmund sez:
# On Tue, 12 Nov 1996, Curt Sampson wrote:
# > On Mon, 11 Nov 1996, Greg Hudson wrote:
# > 
# > > At least one core team member (mycroft) has expressed support for the
# > > idea that the kernel should always keep the symbol table in core...
# > 
# > Ouch! My Sun 3/50 would not like that in the slightest.

Has anyone considered the possibility of keeping the symbol table in core
just long enough to write it out to disk just before the security level
switches from 0 to 1 (i.e. once /etc/rc exits)?

Of course, if you didn't have the disk space for this, it would fail, but
realistically speaking, how large is the symbol table for the kernel?

Assume a maximum of 20 bytes per structure, and 4096 symbols, just for grins.
That's 81920 (80 Kilo) bytes of space.  That's _not_ that much space to
worry about on disk.

Anyway, just as the kernel finishes doing its thing it writes it out
to /netbsd.nlist or something.

Or, how about this:

	- kernel boots, comes up with symtab loaded
	- once minimal functionality is achieved (far enough to get to
	  single-user mode), the kernel _immediately_ pages the symbol
	  table out to swap space and frees that memory.
	- /etc/rc calls a program to extricate the symbol table from
	  swap space into /netbsd.nlist

This, of course, fails if you have insufficient (or non-existent) swap
space (or disk space on output, but if root is that full something's
wrong anyway).

Thoughts?  Comments?

...or am I missing something here? [TM]

				--*greywolf;