'386 Unix Wars

[Dick Dunn]

larry at nstar.rn.com (Larry Snyder) writes:
> tneff at bfmny0.BFM.COM (Tom Neff) writes:
...
> >Anyway, memory is so damn cheap these days...
...
> My beef is that I can only put 12 megs in my machine without causing
> problems with the multiport boards ...

Larry's problem is one of a larger class, which says that memory upgrades
are not necessarily cheap and trivial.  While it's true that you can almost
go to the corner grocery and buy SIMMs at $50/Mb, come home, plug 'em in to
a modern motherboard, fire up and go, there are cases where "just add more
memory" doesn't work now, and more cases where it won't work in the future:
	- Larry's example--memory-I/O problems limits max memory.
	- Common SX boards are limited to 8 Mb.  That may not seem too bad
	  now, but (a) it's a hard limit for those boards and (b) there's
	  no indication that kernel bloat is slowing.
	- Large installations:  It may be no big deal to add 4 Mb ($200)
	  each to a few machines...but what if you've got a thousand
	  machines in the field to upgrade?
	- Memory for older machines/motherboards can be very expensive.
	  The wonder of memory < $50/Mb becomes less wondrous if you have
	  to spend $1000 on a new motherboard to get it.
	- Memory-intensive applications:  If you're using your machines for
	  something real, and it happens to need a lot of memory, sucking
	  up that memory for the kernel may either push you into heavy
	  paging or reduce the size of the largest problem you can solve.
	  The incremental cost of adding a little waste to the kernel can
	  be very high.

But perhaps the reason I find the ever-increasing kernel size so galling is
not so much that it creates certain problems as that it's entirely un-
necessary...
-- 
Dick Dunn     rcd at ico.isc.com -or- ico!rcd       Boulder, CO   (303)449-2870
   ...Mr. Natural says, "Use the right tool for the job."