[Kernel] Redpill Efficient | Blazing Fast | Remarkably Responsive (12.12.2012)

[spline]

Brauch man einen custom Kernel oder nicht?
Auf jeden Fall macht dieser hier einiges her.

---

Update 12.12.2012
V 0.10
http://www.pocketpc.ch/samsung-galax...ml#post1599889

Update 02.12.2012
V0.9
http://www.pocketpc.ch/samsung-galax...ml#post1589190

Update 20.11.2012
V0.6
http://www.pocketpc.ch/samsung-galax...ml#post1581686

---

http://forum.xda-developers.com/show....php?t=1975938

Download
RedPill Stable Revision: Click to download RedPill_Rev0.5_CWM.zip

RedPill_Rev0.5_CWM.zip

[FONT=verdana]Highlights
[/FONT]

• Ramfs from Stock Kernel + Busybox and its various functions in /sbin
• Based on Samsung Galaxy Note 2 N7100 Source Drop Release 1
• Included patches for performance, stability and battery life
• Init.d support
• SetCPU, ROM Toolbox and Voltage Control Support
• No CPU & GPU Overclocking and Undervolting
• Enable or Disable File Syncing
  (fsync disabled by default as I've tweaked the system for optimum latency designed for Flash Storage)
• CFS Autogroup by Mike Galbraith Enabled
• CPU Topology and Sched_MC enabled
• AFTR and LPA enabled
• A lot of tunables via sysfs included (Use ROM Toolbox or similar Utility to easily change it)
• Git Implementation of SHA-1 for 12% faster boot time
• CPU set at 1.6Ghz at Boot for faster boot time (Thanks to Imoseyon)
• LZO compressed kernel using optimized values for the size of the kernel for faster boot time
• CIFS Support (cifs.ko located in /lib/modules) | Tweaked Ext4 Filesystem
  (Patches + Mount Options + Tweaked IO Schedulers leaning towards latency for Flash Storage)
• SIO | noop | deadline (tweaked for better latency and balanced throughput for Mobile NAND based devices)
• Tweaked the mount options for Ext4 to adapt to the current focus on latency
• pegasusq | conservative CPU Governors
• WiFi Multicast Blocked (Thanks to Entropy512)
• Black Crush fix (Thanks to AndreiLux)
• Using Toolchain 4.6.3 from ezterry + compiler optimizations specific for that version (Thanks ezterry)

Power Saving Features:

• AFTR + LPA enabled
• sched_mc enabled (set at 2 by default)
• ARM CPU Topology enabled
• No HZ enabled

Boot Time Features:

• Used git Implem of SHA-1 for 12% Boot time improvement
• Added [PATCH] arm: remove "optimized" SHA1 routines by Linus Torvalds
• Added [PATCH] arm: remove stale export of 'sha_transform' Linus Torvalds
• Set Clockspeed at 1.6Ghz at Boot time to ensure all critical tasks have enough power to perform them while
  the governors/maxfreq isn't set yet. (Thanks to Imoseyon for this hack)

CPU Features

• No CPU Overclocking or Undervolting
• No GPU Overclocking or Undervolting
• pegasusq | conservative CPU Governors pegasusq set as default

Filesystem Features (Currently Supported)

• ExFat (Using Proprietary Samsung Modules)
• Fat32
• Ext2/3/4
• CIFS

I/O Schedulers

• SIO (2012 0.2 version) (tweaked for Flash) Set as Default
• deadline (tweaked for Flash)
• noop
• Tweaked values in deadline and SIO I/O scheduler to give better than average throughput while attemting to improve latency (if not more aggressive). Based on our initial (indicative, not conclusive... yet) testing, having these in line with the vm dirty, expire, writeback values + ext4 mount options to "schedule" write outs as fast as the system can handle it works quite well to balance throughput with latency expected in a mobile device. We took the big picture view and tested our tweaks instead of copy pasting random "known" good values and challenged some assumptions even we had at the start. The result is a mobile device tuned for average throughput and balanced battery life with good latency (not the lowest latency out there, but given the benefits of better I/O throughput and potential battery savings + extending the lifetime of NAND Based storage I think it was a compromise worth taking) I could have tweaked it for extremely great throughput and battery savings but that wouldn't be fun to use.I hate micro-lags myself.
• Our Test Data regarding I/O schedulers and Kernel Tweaks can be found HERE.
• The blog post that describes what we are trying to do can be found on my blog post HERE. (Thanks to s2d4)

Memory Features

• Custom minfree values tweaked for 2GB RAM
• Tweaked vm values in sysctl that's optimized for latency

Display Features

• Black Crush fix (Thanks to AndreiLux)
• Stock mdnie values for more vivid details on the current generation of AMOLED Displays

Performance Patches Enabled

Mike Galbraith's Ultimate CFS Performance Patch (CFS Autogroup)
*More info on this here: https://lkml.org/lkml/2010/11/20/91

Added the CGroup Patch:*Added [PATCH] cgroup: Provides a way of tasks grouping by timer slack value
by Kirill A. Shutemov Based on patch by Jacob Pan. Introduces per cgroup timer slack value
which will override the default timer slack value once a task is attached to a cgroup. It's useful in mobile devices where
certain background apps are attached to a cgroup and minimum wakeups are desired.

Experimental Latency Related Patches

• Enabled "fsync disabled"
  (can be disabled by doing an echo "1" > /sys/module/sync/parameters/fsync_disabled
  in a terminal emulator or as a script for gscript or scriptmanager) (Thanks to Ezekeel)
  Using the tweaked Ext4 filesystem + scheduler and mount options leaning towards latency + vm values in the kernel
  makes the most of the speed of Flash Storage based devices. At the speed at which the data is written to and from the kernel
  to the Fast Storage devices, you would only lose up to 1 second worth of data at most IF the kernel crashes.
  I don't plan on that happening so I enabled the system to get the maximum possible performance in this area.
• Tweaked Ext4 Filesystem (Patches + Mount Options + Tweaked IO Schedulers leaning towards latency)
• SIO | noop | deadline (tweaked for better latency and balanced throughput for Mobile NAND based devices)
• Tweaked the mount options for Ext4 to adapt to the current focus on latency

3rd Party Kernel Apps Support

• ROM Toolbox by jrummy
• SetCPU by michaelhuang
• Voltage Control by xan

	I/O Scheduler Test Results
	Test Partition is /data formatted as ext4 and mounted with the following options: nosuid, nodev, noatime, commit=30, barrier=1, data=ordered, noauto_da_alloc, discard
	Test Data for AndroBench is 16MB Read, 1MB Write Buffer Size for Sequential and Random Access is SEQ=256kb RND=4kb SQLite Transactions=100
	Testing is done simply by changing I/O scheduler using Voltage Control to ensure no init.d scripts overwrite default settings in kernel then AndroBench is run once per I/O scheduler Change. Only 1st run is recorded for now. Testing was done while no other heavy I/O Operations were being run (i.e. listening to music or watching a 1080p MKV. These results WILL vary if those are taken into account. (which will be another Spreadsheet when I have the time)
		I/O Schedulers (RedPill defaults)			Hit the source to see which defaults are used
	AndroBench Storage Benchmarking Tool I/O Variables	deadline	SIO (2012)	noop	V(R)	bfq (v3r1)
	Sequential Read (MB/s)	49.14	49.37	47.63	48.75	47.74
	Random Read (IOPS)	2616.5	2546.03	2429.11	2558.53	2282.68
	Sequential Write (MB/s)	2.52	6.06	2.47	3.83	0.88
	Random Write (IOPS)	60.71	66.41	62.36	60.14	34.28
	SQLite Insert (TPS)	23.99	17.75	22.73	21.28	21.34
	SQLite Update (TPS)	26.52	25.1	20.86	19.23	21.33
	SQLite Delete (TPS)	24.28	23.4	26.8	22.1	21.41
	Browser (msec)	402.75	383	507.5	228.75	330.5
	Market (msec)	799.5	463	725.5	725	516
	Camera (msec)	566.75	355.75	372.25	395.75	803.75
	Camcorder (msec)	2156.5	2052.75	1695.75	1899.75	1856
	While Running a 320kbps mp3 file (low I/O load)
	AndroBench Storage Benchmarking Tool I/O Variables	deadline	SIO (2012)	noop	V(R)	bfq (v3r1)
	Sequential Read (MB/s)	46.14	45.42	45.62	48.12	46.19
	Random Read (IOPS)	2403.71	2393.65	2390.16	2472.08	2335.69
	Sequential Write (MB/s)	3.27	3.26	2.5	2.37	0.51
	Random Write (IOPS)	58.85	56.27	69.74	63.43	34.81
	SQLite Insert (TPS)	30.7	28.97	30.47	24.83	18.78
	SQLite Update (TPS)	25.64	19.74	33.54	20.21	20.87
	SQLite Delete (TPS)	26.22	25.94	29.85	22.29	16.5
	Browser (msec)	444.5	309.75	281.5	283.75	322.25
	Market (msec)	776.75	1038	661.5	951.5	899.75
	Camera (msec)	557.75	444	474.5	535.5	410.5
	Camcorder (msec)	1705.25	2048.75	1751.5	2166.75	1765

[spline]

Ach, falls man wieder zurück will, hab ich hier den Stock Samsung Kernel aus der LJ3.

[Gedel85]

Tönt ja eigentlich sehr vielversprechend... Aber braucht man wirklich einen Custom-Kernel? Hat schon jemand Erfahrungen damit gesammelt?

[spline]

Ich hab ihn seit Gestern drauf.
Einen "wow" Effekt kann ich nicht feststellen.
Da aber die Unterschiede von Custom- zu Stockkernel heutzutage nicht mehr so gross sind, hab ich das auch nicht Erwartet.

Alles ist schnell, flutscht und wegen dem Akku brauche ich noch Zeit.

Gesendet von meinem GT-N7100 mit Tapatalk 2

[Gedel85]

Positiver "Erste-Lese-Eindruck" vorerst bestätigt... Bin gespannt wie sich der Akku verhält. Wenn du da auch nichts Negatives zu berichten weisst, schau ich dann auch mal...

[Jan Badertscher]

Die CIFS und exFAT kernel module sind interessant. Die anderen Tweaks spürt man wohl nicht sonderlich oder? Bin mal gespannt. Ist doch der MIUI Port auf XDA für das N7100 zwar cool aber säuft selbst mit dem Stock-Kernel gefühlt etwas mehr Akku.

[spline]

Naja, der i/o ist schon höher mit dem SIO Scheduler.
Das merkt man auch am Benchmark.

Mir kommt es bis jetzt so vor, als brauch der Kernel einen kleinen, minimalen Tick mehr Akku.

[DakiX]

Wuhu wird getestet.

[Gedel85]

Bin gespannt auf deine Erfahrungswerte du Freak...

[MikeFRG]

Habe ich seit gestern am Start, im Vergleich zu Stock marginal wenn ich die Geschwindigkeit betrache und was Spline schon
schrieb ist der Verbrauch wohl ein bischen höher.

[DakiX]

Naja auf dem Note II braucht man keine weiteren Tweaks wegen der Leistung sondern Akku wäre besser OBWOHL eigentlich GARKEIN custom Kernel nötig ist auf dem Boliden.

[Gedel85]

Hab ich mir doch fast schon gedacht... Dann bleib ich direkt mal brav auf Stock und spar mir die Kernel-Flasherei...

[DakiX]

100% > 89% = 1 Stunde 33 Minuten Display-on
hochgerechnet 13h - mal schauen

[spline]

Hab ihn nun über einen Tag drauf und bin bei 56% Akku übrig.

WLAN 26 Stunden daueraktiv (mit push).
Bildschirm 2,5 h

Gesendet von meinem GT-N7100 mit Tapatalk 2

[spline]

Gesendet von meinem GT-N7100 mit Tapatalk 2

[DakiX]

Der Kernel ist top werde aber morgen berichten keine Zeit jetzt (eher Lust)

Gesendet von meinem Nexus 7 mit Tapatalk 2

[Milchbeck]

Kann es sein das es Internet Empfangs Abbrüche mit dem Kernel passieren können? Habe nämlich seit dem ich ihn benutze viele kleine Aussetzer aber nur Mobile Daten WLAN läuft super

Send with Note 2
Wanam 1.5
Motomoto00 by Raubkatze

[spline]

Seit über 3 Tagen drauf.
72h WLAN und Push aktiv.

Keine Probleme.

Gesendet von meinem GT-N7100 mit Tapatalk 2

[DakiX]

Hier auch alles super. Hält gefühlsmässig vom Akku her länger und Performance nach wie vor 1A.

[DakiX]

Eben an die Dose gesteckt mit 7%.

100% -> Kernel geflasht und normal benutzt etc. Ergebnis: 3 Tage 7 Stunden und 38 Minuten davon waren 7 Stunden und 24 Minuten Display-on. Nice werde ihn beim einten Note behalte denn beim zweiten ist Perseus drauf und aktuell noch 41% -> 4 Tage, 12 Stunden und 16 Minuten, davon 4 Stunden und 53 Minuten Display-on.

Ich liebe das Note 2!!