Re: zram Usage as Default in Debian (?)
On 15/01/2012 19:52, Rainer Dorsch wrote:
On one system, that is exactly, that is what I observe.
Ok, i suspected that.
I've haven't yet found the time to test zram and surely will do. But i
also admit that the more i think the more seems to me the wrong way to
solve the responsiveness problem. I would expect a kernel setting that
can tweak how it use the swap area.
I just wanted to say to the kernel: cache is less important than
processes; swapout processes only when the physical ram space is really
ended or it's next to it.
The vm.swappiness variabile seems to be the parameter nearest to that
but in my experiments the perceived effect in day to day usage is null.
In other words, setting it to zero, doesn't reduce the swap so much and
after hours of use i still have 50-100 MB used swap space (belonging to
processess that i will probably use and that will cause strange lags
documented only by the hdd led).
The more detailed document i've found so far is that:
But i admit that there are things that i've not fully understood yet and
apart from swappiness i have to test the other options yet.
Zram looks like a suboptimal solution to me, even if it can work in some
scenarios like mine and yours. I feel it more like a workaround that
come at the price of lost ram space and computational power (could be
Since i was not able to find a clear kernel parameter to tune swap usage
like i want and since my ram is usually filled below 50%, i'm now using
my laptop without swap space. And clearly, from the responsiveness POV,
that works. ;-)
I've prepared a 1 GB file formatted with mkswap to use if needed.
The other system is a low end system with 512 MB RAM and it is running KDE4.
With zram the system is much more responsive than just with HDD swap space.
Mmmh, my feelings is that, actually, zram make sense only if you have
plenty of free ram. In low ram PC like yours i expect every bit of ram
should be precious and that you could easily end in what others have
already said: more negative side effects than improvements. It would be
useful only if the swapout data were highly compressible, which i don't
know and that's why i will make some test when i'll have more spare