maemo.org - Talk - [ANNOUNCE] CSSU-thumb thread

You can't use UBIFS as swap or format it in any way. N900 bootloader (forget it's exact name) boots /sbin/preinit from UBIFS. Changing that requires bootloader modification and flashing it into MTD0 (or MTD1 - I don't remember for a moment).

Besides of that it has no big sense - swap functions in kernel use a big blocks for swapping, so there is no advantage in OneNAND UBIFS vs microSD. It is a little faster, of course, but it would be interesting to test Class 10 and see.

Megaltariak

2012-06-15 11:42

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Didn't even know that they did that for the N9.

Now the problem is either hacking NOLO to boot on another storage device (would be really difficult because NOLO is closed source) or resize/create new partitions on the 256Mo chip (would be trivial with standard tools if it was recognized as a standard storage device like EMMC, but that is not the case if I recall correctly)

hxka	2012-06-15 12:47

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

That's a huge offtopic, but…
It is not NOLO mounts ubifs and runs /sbin/preinit. It is KERNEL. And we can recompile it or use u-boot to change it's cmdline to mount /dev/mmcblk0p2 as /. So, NOLO doesn't needed to be changed.

Obviously, swap on UBIFS with compression is bad choice. But mtdswap or even swap on uncompressed UBIFS is interesting idea to try.
Anover idea is to optify some more libs and other stuff and use uncompressed UBIFS as rootfs.

@Megaltariak, partitions on NAND are either compiled in kernel or specified in it's cmdline.

freemangordon

2012-06-15 17:45

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

New version (21.2011.38-1Tmaemo4-thumb7) is available, check the first post for changelog

doctorwhiz

2012-06-15 18:43

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

i installed successful; tanx

Fabry

2012-06-15 19:09

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by Megaltariak (Post 1222222)

I believe that choosing UBIFS instead of standard ext3 was a really bad decision from Nokia

UbiFS is a MUST for Nand.

Nand is a raw flash chip (MTD) and not a generic block device so we need a software layer which handles directly a MTD and implements: tracking NAND flash bad blocks and providing wear leveling.
UbiFs / JFFS2 / LogFS are this kind of software layer

eMMC or SD Card have a custom hardware chip which handles the flash chips, so they are generic block devices that can be formatted with Ext2/3/4 or other FS

Alfred

2012-06-15 21:32

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

had to reflash after upgrade...

szopin

2012-06-16 00:43

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by freemangordon (Post 1222167)

You're missing my point, it is not about the writes at all, rootfs is much faster on writes, but that does not matter when we're talking about executable code ;). Is is all about the read operations, and there is where rootfs is not only slower, but consumes 90% CPU time to achieve that slowness. BTW the reason i choose 95MB file to do the read test was to get rid of cache/buffers/whatever effects on the speed.

Also have in mind that there is no performance penalty for doing random reads on flash (i.e. /opt) it is only write operations that are affected.

Nah, there is no such thing like compression chip, it is the standard ubifs de/compression and it uses CPU. I definitely say it is slower, as my test ( the "time cat" one) shows the maximum reading speed achieved when there is noone else using the CPU. Imagine what happens with that speed when there are several processes using the CPU.

The case is closed for me, expect a new update (hopefully today) which moves Qt back to /opt.

Sorry for corpospeak, but that clarifies, thanks!

seanmcken

2012-06-16 03:18

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

wats difference between power kernel and this custom cssu kernel??

hxka	2012-06-16 04:43

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Did you try to read first post?

szopin

2012-06-16 06:14

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Ok, just thinking out loud, pls correct me if I'm wrong, but if thumb2 saves 3-4mb out of 14 on average... we should be able to free extra 45 out of 150mb in rootfs that are there by default(add some optifying for good measure). Is killing compression an option? Should give speed/battery(no CPU overkill) gain for most (I'll stick to having libs there, but standard user...). Then again the 150mb that are there are already compressed, so would expect them uncompressed to take over 200, thumb2 could help this???

Hurrian

2012-06-16 07:23

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by szopin (Post 1222594)

Yup, UBIFS with no compression is an option.

Anyone want to try to debootstrap a Thumb2 rootfs to flash with flasher (or alternatively, backup then generate an uncompressed fs and untar it back)

freemangordon

2012-06-16 07:30

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by szopin (Post 1222594)

You may experiment a bit by creating a directory in / , lets name it rootfs_uc. After that do chattr -c /rootfs_uc, that command disables compression for a specific node. Once compression for rootfs_uc is disabled, you can start moving and symlinking binaries from /usr to /rootfs_uc.

EDIT:
@Hurrian - there is no need to uncompress the whole rootfs, we can do it on-the-fly only for what is really needed, see the above

lma	2012-06-16 09:07

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by szopin (Post 1222594)

Is killing compression an option? Should give speed/battery(no CPU overkill) gain for most

That would avoid the CPU overhead for compression, but incur more physical I/O (which in itself uses some CPU anyway) for the same amount of data. It should be benchmarked carefully to determine whether the tradeoff is really worth it, both in terms of performance and energy consumption.

hxka	2012-06-16 10:05

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

For now, I have

Code:

Nokia-N900:~# du -xhs /

345.6M  /

So, without optifying uncompressing whole rootfs is not an option.

hxka	2012-06-16 10:21

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by freemangordon (Post 1222618)

chattr -c /rootfs_uc

Hmm, it's from e2fsprogs package, are you sure it will work with ubifs?

freemangordon

2012-06-16 10:36

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by hxka (Post 1222684)

Hmm, it's from e2fsprogs package, are you sure it will work with ubifs?

yeah, I tried it, by creating a directory under /root, setting -c flag on it and copying a big file into it. doing the same filecopy in a different directory and checking the remaining space with df -h

hxka	2012-06-16 10:39

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Code:

Nokia-N900:~# mkdir /uc

Nokia-N900:~# chattr -c /uc

Nokia-N900:~# df -h /

Filesystem                Size      Used Available Use% Mounted on

rootfs                  227.9M    179.1M     48.7M  79% /

Nokia-N900:~# dd if=/dev/zero of=/10Mfile bs=1K count=10K

10240+0 records in

10240+0 records out

10485760 bytes (10.0MB) copied, 0.169525 seconds, 59.0MB/s

Nokia-N900:~# df -h /

Filesystem                Size      Used Available Use% Mounted on

rootfs                  227.9M    180.7M     47.2M  79% /

Nokia-N900:~# dd if=/dev/zero of=/uc/10Mfile bs=1K count=10K

10240+0 records in

10240+0 records out

10485760 bytes (10.0MB) copied, 0.195251 seconds, 51.2MB/s

Nokia-N900:~# df -h /

Filesystem                Size      Used Available Use% Mounted on

rootfs                  227.9M    187.4M     40.5M  82% /

Nokia-N900:~#

Hmm, strange. Looks like it works, but used space increased by ~7M, not 10M.

freemangordon

2012-06-16 11:42

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Have in mind used/free space on rootfs is an estimate AIUI

Hurrian

2012-06-16 12:01

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by hxka (Post 1222674)

For now, I have

Code:

Nokia-N900:~# du -xhs /

345.6M  /

So, without optifying uncompressing whole rootfs is not an option.

Hmm, it seems that this is true.

Has anybody decided whether the slightly slower program load times on eMMC justifies lower CPU load?

(it's not exactly reading/writing enormous amounts of data, so IO CPU use wouldnt really choke the CPU)

lma	2012-06-16 12:47

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by hxka (Post 1222693)

Code:

Nokia-N900:~# dd if=/dev/zero of=/10Mfile bs=1K count=10K

10485760 bytes (10.0MB) copied, 0.169525 seconds, 59.0MB/s

Nokia-N900:~# dd if=/dev/zero of=/uc/10Mfile bs=1K count=10K

10485760 bytes (10.0MB) copied, 0.195251 seconds, 51.2MB/s

Interesting that /uc is ~15% slower. Is that reproducibly so?

hxka	2012-06-16 13:09

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by lma (Post 1222737)

Interesting that /uc is ~15% slower. Is that reproducibly so?

That's obviously because of zeroes are too easy to compress, and compressing isn't a bottleneck. Remember that dd calculates speed of writing data to file, not an actual speed of writing (un)compressed data to flash.
For 'real' usage that definitely won't reproduce.
But I'm lacking of fast source of pseudorandom data to test it, /dev/urandom is VERY slow on N900 (~1.6MBps on my N900 OC to 900MHz).

szopin

2012-06-16 14:53

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by hxka (Post 1222674)

For now, I have

Code:

Nokia-N900:~# du -xhs /

345.6M  /

So, without optifying uncompressing whole rootfs is not an option.

This is actually great news and an option. 30% of 345 is 103mb, so less than 256. The more optimistic view of 40% leaves around 50 mb of rootfs. Thumb2 could bring yet unknown performance (both using uncompressed rootfs and smaller code footprint) and significant battery saving (no CPU overhead on any/most system processes). This thread is pure gold, fingers crossed.

jonwil

2012-06-16 14:53

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

The code in https://garage.maemo.org/plugins/scm.../?root=browser should be matching source code for tablet-browser-daemon (i.e. browserd), there is also source at https://garage.maemo.org/plugins/scm.../?root=browser for browser-neteal.

The browser UI itself is not open source unfortunately :(

hxka	2012-06-16 17:15

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

@szopin, not all of files in rootfs are ELFs.
It's bit tricky to count size of all ELFs in rootfs, I've ended up with this:

Code:

Nokia-N900:~# cat iself.sh 

#!/bin/sh

file "$1" | grep ": ELF" 2>&1 >/dev/null && echo "$1"

Nokia-N900:~# du -ch $(find / -xdev -type f -exec sh iself.sh {} \;) | tail -1

191.2M        total

And 30-40% is a reduction of code size, not resulting code size.
So, 345.6−191.2⋅0.4=269.12. Too much, even if not to take into account that not all of them are open source and can be recompiled, that partition size is 251.25 MB, not 256 and that filesystem iteslf require some space to store inodes and other stuff.

Mohammed Muid

2012-06-16 17:17

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by szopin (Post 1222793)

BusyBox v1.20.1 (Debian 1.20.1power1) built-in shell (ash)
Enter 'help' for a list of built-in commands.

~ $ root

BusyBox v1.20.1 (Debian 1.20.1power1) built-in shell (ash)
Enter 'help' for a list of built-in commands.

Nokia-N900:~# du -xhs /
298.1M /
Nokia-N900:~#
i got this. i have no idea what is this. i just installed it. absolutely no clue. just understood that it will make N900 a little faster and apps will eat less RAM

shabmanu

2012-06-16 18:10

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

just now i updated from tumb6 to tump9 but rootfs is still same why?

freemangordon

2012-06-16 19:18

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Guys, I would really love if you stop measuring WRITE speed, what matters is READ speed, as that is what is applicable for executables :P

Also we don't really need to move all of the executable code in an uncompressed directory (there is no enough space), just the one that is frequently used. Even if not thumb-compiled.

hxka	2012-06-16 20:56

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

If somebody interested, on a pure PR1.3.1 (only rootsh, file and busybox-power installed) there are 263.1MB of files on rootfs and 149.6 MB of them are binaries.

Alfred

2012-06-17 15:24

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

I have updated to the latest version, and have been experiencing a little slowdown...

freemangordon

2012-06-17 15:39

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by Alfred (Post 1223247)

I have updated to the latest version, and have been experiencing a little slowdown...

Would you elaborate on that? What exactly is slower and how did you measure the slowness. Which version did you upgrade from?

Cow	2012-06-18 02:07

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

im a newbie here... how do i overclock my n900 without kernel power? thx in advanced..

geneven

2012-06-18 03:35

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

If you installed this, you have kernel power.

menghao1993

2012-06-18 12:08

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

unable to update this ..
package corrupted..
any idea? :o

Cow	2012-06-18 14:14

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

will i need to install kernel power setting? in order to use kernel-config in x-terminal ?

geneven

2012-06-18 14:28

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by Cow (Post 1223698)

will i need to install kernel power setting? in order to use kernel-config in x-terminal ?

If it isn't installed, you probably need to install it. I already had it installed. As root, type kernel-config show to get a clue.

MastaG

2012-06-18 14:34

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Yo Gordon Freeman!
My man! <Denzel Washington-voice>
It works great :)
I run more apps now (like 20 open windows) without the phone getting slow.
Great work! I hope this will soon hit the CSSU-testing with a complete recompile of everything:P

Btw: Do you think compiling your 720p DSP codec will benefit from this?

ivgalvez

2012-06-18 15:09

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

A good proof of concept, apart from the already mentioned MicroB (browserd always loaded in memory) and Firefox, could be building Python with Thumb2.

hxka	2012-06-18 15:34

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by MastaG (Post 1223707)

DSP is very different processor and it has nothing to do with CPU.

Alfred

2012-06-18 15:42

Re: [ANNOUNCE] CSSU-thumb thread - stable Thumb2 on N900

Quote:

Originally Posted by freemangordon (Post 1223253)

Would you elaborate on that? What exactly is slower and how did you measure the slowness. Which version did you upgrade from?

Okay, let's say there were 3 versions of Thumb2 CSSU so far. when i installed 1st when i already had CSSU-T a few things got messed up (like i had only 4 desktops), but i didn't pay much attentions to that, because phone got responsiver and fast in overall. Then i updated and n900 bricked, so i did a clean install of the 2. version, without preinstalling CSSU-Testing. That went all well, and even "fixed" desktops and other little bugs. After that i updated to the last version, and phone slowed down say... 10 to 20 percent. But i will do a reflash today, and do another clean install of the last version available. Therefore i have a qusetion, do You think a eMMC reflash in this case is usefull, or shud i just reflash with a COMBINED image?

Keep in mind i didn't overclock in any of the cases...

All times are GMT. The time now is 00:02.

Page 4 of 57

Show 40 post(s) from this thread on one page

vBulletin® Version 3.8.8