Thread: Boost N900! :)
View Single Post
Posts: 436 | Thanked: 406 times | Joined on Jan 2010
#265
Originally Posted by tokag View Post
@ SavageD, or whomever might know...

first off i would like to say a big THANK YOU!!! to SavageD for all your work and experimenting in putting these tweaks together.

i took a peak inside your TuneN900.sh script to see how your system tweaks varied from my own. i prefer to hard-code my tweaks directly into my rcS and rcS-late files rather than running a modifying script at start (i don't use Swappolube for this reason). i noticed some interesting discrepancies between your ideal setting and what i thought were ideal setting, and i was wondering if you might be able to shed some light on why you chose the values you did.


echo 6 > /proc/sys/vm/page-cluster

i currently am running with this value as 3. some believe 1 is a good value while other's argue that the leaving this @ 5 (nokia default) is better. i landed somewhere in the middle and it seems to be working well for me. what effects did you notice while experimenting with this value?

echo 900 > /proc/sys/vm/dirty_expire_centisecs
echo 900 > /proc/sys/vm/dirty_writeback_centisecs

i current have both these values set to 0 (disable). i believe the nokia default was 500 or something in that neighborhood. is it actually beneficial to allow these daemons to run (just less frequently than nokia's default)?

echo 40 > /proc/sys/vm/vfs_cache_pressure

i had originally left this @ it's default value of 100, but decided to give it a go @ 40. having changed multiple values it's impossible for me to be able to tell what exactly the effects of this are on my system. so i ask you, whom i assume experimented with control tests, how much exactly does this setting effect the devices performance?

echo 60 > /proc/sys/vm/dirty_ratio

again i cannot remember the default value for this, but i changed mine to 95 after reading about it in a blog a while back. does this value work better after being set lower? i realize setting it too high risks of running the device out of memory, but setting it too low causes unnecessary writes as it starts dumping dirty writes too often. obviously you feel 60 is a good value for this, but i am curious how much testing (and what results) you did before arriving at this conclusion.

well i think that's it for now. Thanks again, and keep hacking!

http://www.knownokia.ca/2010/08/n900...-and-more.html
(original blog post i used for modifying vm setting)


don't want to go into techincallities and all cause I don't want to make a long post.

echo 6 > /proc/sys/vm/page-cluster

I chose this value simply because having done some research I found that increasing the value actually made their computers faster. The original value was "5" which I thought was good enough though just to make things slightly better I increased it to "6" any higher and it would dig into battery life as it makes swap more aggressive.

echo 900 > /proc/sys/vm/dirty_expire_centisecs
echo 900 > /proc/sys/vm/dirty_writeback_centisecs

After reading this, and many other resources; I decided to make the pdflush daemon check dirty memory less frequently to reduce cpu load when using heavy apps, longer checks intervals means less pause time....Where dirty memory was checked/flushed every 500 milisecs it is now checked/flushed every 900 milisecs.

echo 60 > /proc/sys/vm/dirty_ratio

Actually Changing this value two I saw no increase in anything. I chose the value 60 simply because it was intermediate between 40 and 95. If needs be I would probably change it back to 40 :/.

echo 40 > /proc/sys/vm/vfs_cache_pressure

How should I put this :/...basically I reduced the total memory or caches dedicated to the checking and indexing of the file system to 40%.
 

The Following 2 Users Say Thank You to SavageD For This Useful Post: