> >My understanding is that total virtual memory is essentially equal to
> >physical RAM + swap-size.  Correct?  And, although a swap-size of approx.
> >4x physical memory is recommended, there's no particular condition on the
> >size of the swap paritition.  Correct?
> >
> 
> I don't know if UVM has changed this requirement but, yes, there was a
> lower limit on the amount of swap you had (I suspect UVM has lifted
> this but...).  In older BSD systems you needed to have at least the
> same amount of swap as you had physical RAM because each memory page
> needed to be backed by a swap page.  The old rule of thumb was at
> least twice the physical RAM size was a good swap size to give you
> some pages for data.

Hm.  Ouch.  That's what I was afraid of.  I really didn't want to
repartition my drive just because I'm adding memory, though.  (I thought
that, even pre-UVM, under 1.3, there were options about explicitly
disabling the swap---e.g., for diskless workstations.)

My hope was that that meant that I could have arbitrarily small swap
space.  (In this case, ``small'' would be about 4/5ths of my actual
memory.)

From the sound of it, you aren't sure what the current kernel requires,
though.  Is that correct?


> In any event if you expect to analyse kernel core dumps then you need
> a single swap partition big enough to hold the memory image.

Hm.  I don't expect to do that.  (^&  (Other than a 1.4 kernel problem
with moderate loads (thus I'm now running a 1.4.1 kernel), NetBSD strikes
me as extremely stable.  I don't run -current, and am not a kernel hacker
at this point.)


Thanks for your help.


  "I probably don't know what I'm talking about."  --rkr@rkr.kcnet.com