[KERNEL] [Y300/G510] Stock Huawei

[Guest]

@fonz93: personally I'd recommend finding the most suitable value by testing.

[Guest ZolaIII]

It did got in some Linux builds far from mainline.

The thing is value of 4 works good & cash is not getting filed up to top bat it rises performance & saves some energy by not doing so often expensive reclaims. Their is still no reason for this to be tunable & not a fixed function value (set to 4) like in mainline. Did some testing naturally. Google didn't do a proper documenting... It's always actually just about finding best compromise setting that will serve purposes. :)

[Guest]

Their is still no reason for this to be tunable & not a fixed function value (set to 4) like in mainline.

 

I'm sorry, but Linux's mainline default value is still "1".

 

See here: https://github.com/ARM-software/linux/blob/master/mm/page_alloc.c (Line: 1702)

 

Now compare Google's commit again: https://www.codeaurora.org/cgit/quic/la/kernel/msm/diff/?h=msm-3.4&id=83de731ffcc6777a33e8a6132c7da8d91faac9ca (They changed it from 1 to 4 - and this only because Android doesn't natively have SWAP).

 

Linux default:

/* Require fewer higher order pages to be free */
min >>= 1;

This value got made tunable through sysfs and is getting set to '4' during boot by a ramdisk script.

Edited July 21, 2014 by Guest

[Guest ZolaIII]

I mean it's not a tunable in mainline bat a fixed function... I know the mainline set value.

[Guest]

Maybe I misunderstood you then. I thought you claimed the mainline's default value was 4.

[Guest]

New TEST BUILD for debugging performance and possible errors: http://www3.zippyshare.com/v/37695080/file.html

 

• kernel got downgraded to 3.4.0 again
• disabled frandom being enabled by default (the driver is still present @ /dev/frandom - it just doesn't automatically replace /dev/random anymore)
• upstreamed the kernel's default /dev/random driver
• enabled CPU frequencies below 245 MHz

 

I'm not completely done yet. Would be nice to know if you experience any regressions or notice any bugs so far.

Edited July 21, 2014 by Guest

[Guest fonz93]

I did some tests, 3D performance increased for about 150-200 points on antutu, database io decreased for about 120-150 (maybe because of the disabled frandom), the overall score is pretty the same (7800-8000 more or less)

[Guest ZolaIII]

What OOM settings are you using Fonz?

& what min_free_order_shift value?

Edited July 21, 2014 by ZolaIII

[Guest fonz93]

Default OOM values (the ones provided by synopsis.rc) and min_free_order_shift=2

The same configuration of this morning tests

Edited July 21, 2014 by fonz93

[Guest ZolaIII]

Try like this:

echo 3584,4864,6144,7866,9832,11799 > /sys/module/lowmemorykiller/parameters/minfree;

& use min_free_order_shift=4 default value.

I am getting strange very hi RAM write performance values in some tests higher then ever (55% higher then in previous build)??? 3D EGLS 2.0 performance is up a little.

Edited July 21, 2014 by ZolaIII

[Guest fonz93]

Your minfree values are near to moddingg33k's values, i'll try them tomorrow morning as i haven't enough battery at the moment

[Guest ZolaIII]

OK! Sorry & good night!

[Guest Picvajzmen]

This new test build for me is fast as hell. Great stuff guys for now. Much quicker then the last one.

[Guest]

If you want to re-enable 'frandom' on this test build you can simply run this command in a console once:

start frandom

[Guest fonz93]

Another change that i noticed was the runtime, with 17-07 and yesterday morning tests, i reached 1005, with yesterday build i reached max 770 and after changing min_free_shift_order=2 to =4( kernel default) this morning runtime decreased to 680-700 (but i think not a big change, the big change is from 17-07 kernel and yesterday's build)

No changes in RAM values

Edited July 22, 2014 by fonz93

[Guest]

I never reached such values for runtime, regardless of the build.

My max'es are between 700-800

Edited July 22, 2014 by Guest

[Guest fonz93]

The values were stable for 3 tests, so i think wasn't "fake", i got 980,1001,1005

 

 

Edit: I tried to get again those scores (i flashed again the 17-07 and restored my yesterday's configuration) and my max now is 750  :huh:

Edited July 22, 2014 by fonz93

[Guest ZolaIII]

I am getting around 720 & up to 750 with last build & nothing changes since one before. Never got dalvik score more than 790 (Y300).

[Guest fonz93]

How could i reach 1000 yesterday? The only changes were related to the kernel and min_free_order_shift

Today i did a wipe dalvik due to an update, i don't know if can changes things

[Guest ZolaIII]

I don't know how or what dose Antutu using to measure dalvik performance bat that not really corresponding well to real usage. Dalvik performance is more bounded to CPU performance all do mem & nand performance also determine it as it's hybrid just in time compiler. On the G510 for instance score is normally up for around 100 points thanks to CPU... I can't explain your scores Fonz & I did actually tune dalvik behavior as much I did know...

[Guest]

At the end it's more important how your experience and impressions look like, rather than a high number on any benchmark.

I've had builds scoring very high, but they were rubbish for general all-day-usage.

[Guest SH3H1]

Is underclock frequencies really help in screen on usage?

Btw new kernel is great in daily multitasking :)

[Guest]

I'm not a big fan of making use of those lower frequencies. I've only enabled them in case someone want's to test them and because it doesn't affect anything else as long as default settings don't get changed.

[Guest]

How could i reach 1000 yesterday?

 

When your scores dropped from 1000 to 700 ... did you use during those runs? If not try to "enable" it and see if something changes.

[Guest pbr1111]

With the latest test build: