Subject: Re: Swap sizes vs. physical memory.
To: Claude Marinier <claude.marinier@dreo.dnd.ca>
From: David Brownlee <abs@anim.dreamworks.com>
List: netbsd-help
Date: 09/09/1999 09:25:52
Could someone add this to the FAQ? - thanks :)
On Thu, 9 Sep 1999, Claude Marinier wrote:
> In summary,
>
> 1) you can run with no swap space at all
>
> 2) if you want a full core dump after a crash, swap space must
> be as large as physical memory (or a bit larger?)
>
I believe the first swap partition must be slightly larger than
physical memory (the dump code does not know how to split across
multiple partitions).
> 3) the size of swap is determined by your work load: active
> processes, dormant processes, transients, ...
>
> 4) four times physical memory is a relic of the past, with
> today's typically large amount of physical memory, the
> recommendation should be somewhere between 1.0 and 2.0 times
> physical memory (or should that be 1.x and 2.0 ?)
>
That seems very reasonable :)
> NetBSD appears to allocate swap space using the 'deferred' or 'lazy' mode.
> How does NetBSD actually allocate swap space? Could a process be killed
> because of insufficient swap space?
>
That is correct - NetBSD does not preallocate swap space based on
programs' current data usage. In a preallocation model:
virtual memory = min(physical memory,swap).
In the NetBSD model:
virtual memory = physical memory + swap.
This means you have more virtual memory available, but is it
possible to malloc memory correctly and then fail when you come
to use it.
In either model a process can be killed when you run out of VM.
> I do not expect a long dissertation in response; rather a short
> explanation (if this is possible) or a pointer to a document on the net
> somewhere. I note that no one made reference to such a document in this
> thread. Is there such a thing?
We should probably ensure it gets added :)
David/absolute
-=- Trust is a non-renewable resource -=-