Ideal Swap Partition Size

[Patrick O'Callaghan]

On Thu, Jan 22, 2009 at 12:05 PM, Tim <ignored_mailbox at yahoo.com.au> wrote:
> Tim:
>>> It seems strange to think that a system will swap just because
>>> there's swap space available.  Surely, it's only going to swap when
>>> it needs to, and you'd be faced with operating and swapping, or being
>>> unable to swap and unable to operate once you reach that stage.
>
> Patrick O'Callaghan:
>> True, but in some systems -- especially those with realtime
>> requirements -- it can actually be preferable to fail than to go slow
>> (due to what we oldtimers call "thrashing").
>
> But it wouldn't be case of your application running or not, it'd be a
> case of something else dying off, as some additional task was started up
> at an inopportune moment.  A Linux OS isn't just one running process,
> even if you only run one application on the box.

Er, I think I knew that :-)

The scenarios in which dying is better than swapping are definitely
not common, but typically a real-time system will be configured to run
only certain processes anyway, i.e. it's not a general-purpose system
that just happens to have some RT app running on it.

Note that this is in no way implies that just having more swap is in
itself going to make the system use it. I was simply putting forward a
situation in which the general rule (more swap is better) might not
apply.

poc