Swap to SD cards: performance and burnout test

[Neil Graham]

On Mon, 2009-12-07 at 19:16 +0100, Sascha Silbe wrote:
> On Mon, Dec 07, 2009 at 03:41:10PM -0200, Emiliano Pastorino wrote:

> > Does anyone come out with a possible test?
> Compilation in general (e.g. Linux kernel or sugar-jhbuild) seems to be 
> quite stressful to SD cards and often consumes a lot of memory, so might 
> be a good benchmark.

I don't think it's terribly useful to test memory consuming
non-interactive tasks.  The important factor with things like
compilation is that it gets the job done,  speed is preferale but not
necessary.   However Delays of the sort that can double the time of a
build operation could quite possibly render a user interface totally
unusable.

The easiest way to test swap performance on working data and a user
interface I would guess would be to run Firefox up to 300 meg. 

Short of running evolution that's the best way to frivolously eat up
your ram. Then just try doing some things. 

I run a swap on my XO, The speed difference while actively swapping
compared to running totally in ram is quite significant.   I'd class it
as barely usable.   What it does give you is the ability to not crash
when you run out of ram, and to do the memory intensive non-interactive
tasks mentioned above.  

I haven't traced it all through but I imagine there is some benefit to
swapping out parts of programs that used memory at boot time and are now
largely idle.  Those items won't swap back in effectively giving you a
little more ram.  Not really enough to have a huge impact though.

What I have found is that the XO has more than enough ram to run all the
lightweight programs it wants, and not early enough to run the
behemoths.  There isn't a lot of middle ground.