I found out that after PR1.3 you can completely turn off swap and left device swapless.



I ran here 3g connection, browser, telepathy and media player. Note: performance have somewhat degraded.

To turn magic on:

Next point is to reduce memory consumption. Any ideas?

Hmm is it me, or browserd uses more memory than PR1.2?

Wow, your lucky to have a functioning device after that o.o...Did you do a restart? Cause I don't think the device would turn on without enough swap...

I think I just came. Not wearing down the swap flash chip is that exciting to me.

Now to figure out a way to make it so that when the N900 actually maxes out the RAM, it enables swap, and uses it to the bare minimum. And then when the need goes away it doesn't just flush swap (seems like that would cause extra writes, unless someone who knows better can correct me on that), but doesn't swap more and keeps stuff out of swap again until absolutely need be.

But either way, this opens up potential for people way better than me to make the N900 perform even better over-all.

no reply. guessing his phone wouldnt work after reboot :P

Chance "swappiness to 1". echo 1 > /proc/sys/vm/swappiness

Is that literally all it takes? Wouldn't that be swappiness 0, not swappiness 1? Since swappiness 0 will still swap when absolutely necessary, right?

(At the end of the day, I am more or less content with swappiness 20, and following the other values that were eventually integrated into swappolube - though I'm currently unsure how I feel about dirty-expire-centisecs and whatever the other similar sounding setting was [the unsureness comes from a not-complete understanding of the technical aspects of how the daemons involved are influenced by those settings, and what they do exactly]. My main desire, and dissatisfaction with the N900's swapping, was that whenever I type a long enough forum post/message, is always, invariably, eventually starts to swap. And because it swaps relatively constantly beyond that point, it causes that lag with the message input, combined with utterly missed characters. I realize part of this is probably a fault of Hildon-Input-Method, not just kernel swapping behavior, but I undeniably noticed that things weren't as bad with swappiness at 20. The point, though, is that I understand that memory and swapping is a complicated enough matter that if I JUST slam swappiness to almost-nil, it won't be good for the device elsewhere. However, if I am able to keep device performance similar as it is already in all my other use cases, I wouldn't mind tweeking parameters to that point. I was experimenting with this before pr. 1.3, now I haven't gotten around to messing with the settings since - especially since I wanted to give it a try and see how it runs currently before going in and mucking about with that values it sets by default at every boot.)

Well, I turned off swap after device is powered on. However, I found out that lowmem settings came from /etc/event.d/rcS-late. I need rescue console to safely test that