Questions re swap and hibernate interaction on 8 gig machine

[Lennart Sorensen]

On Fri, Jun 25, 2010 at 10:18:34AM -0400, Robert Brockway wrote:
> I notice a number of people have mentioned running with no swap.  While  
> that is certainly possible I still argue in favour of having some swap,  
> even if swappiness is set to 0.
>
> My arguments are set out here:
>
> http://www.practicalsysadmin.com/wiki/index.php/Swap
>
> IIRC the system wouldn't swap unless it had _at least_ 2x ram which meant 
> most systems stopped swapping entirely :)  That bug was fixed *fast*.
>
> I'm not a fan of using tmpfs for /tmp.  I think it is a solution looking  
> for a problem.   Essentially I argue that most apps write so little to  
> /tmp that the use of a special case isn't warranted.
>
> Other uses of tmpfs may make a lot of sense though.
>
> The other main reason for swap is the one the OP asked about -  
> hibernation.  A swap partition is needed - you can't swap to a swap file  
> under Linux[1].  In principal you'd need as much swap as ram to guarantee 
> a successful hibernation but it could work with less.  If there aren't  
> sufficient resources the hibernation should fail gracefully.  Personally 
> I use sleep a lot and rarely hibernate.
>
> [1] Last time I checked.  Theoretically it is possible.  I'd love to hear 
> this had been fixed but I haven't seen any reports.

Linux has supported swapfiles for years.  They work fine.  Now I don't
think hibernation works with that, although I am not sure.

Swap itself works great on swapfiles and doesn't even have any significant
performance difference compared to a swap partition, assuming you don't
make a sparse file when creating the swap file (so don't do that).

-- 
Len Sorensen
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists