Re: [ANNOUNCE] COMPCACHE on kernel-power
 

If you do not know what something does, it is the opposite of being polite (i would have said you are a rude a55hole but the forum rules, y'know) to keep asking the same question over and over, right in between a conversation that is held between several other peoples on said $subject, and especially/ if they are technical conversations and you would have very easily found out about $subject yourself.

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

ok i got this .. but it's useful now to install ? or it's pre-release?

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

i got the same result as karam and mohammad
laggy when the swap gets almost full

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

And all of you are using one and the same "magic patches", right? Which were reported by several users(including me) to result in massive lag under memory pressure? Not that I am saying that ramzswap works as it should (notifications are still not working), just wondering.

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

that is true i'm using speedpatch but i have removed it before trying COMPCACHE
TBH after i installed the 'magic patch' COMPCACHE performed very better (at least no much lags)
and you are mistaken the 'magic patch' perferms very well under high memory pressure
But let us not change the subject of the thread bro..

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

The "magic path" is cause of ~50% of semi-unknown problems reported by users. It's damn thing, that it's often mistaken by new users for legitimate one, in line with CSSU, KP and HEN. Does anyone even know, what exactly "magic" is doing, at all?

Sorry for pushing this off-topic further, but I think it's important to process of testing ramzswap. I don't know if reports by used, whose device are "poisoned" by using 'magic' are useful, due to untraceable nature of problems.

How can we know 'magic' is removing itself properly, and don't leave harming remains?

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

I'm using the I/O Interactivity swappolube settings.
And the ohmd/syspart tweaks from here.

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

oh god
Estel u kidding me?
i've seen no one complaining about it
and freemangordon
this magic patch you are talking about
is the BEST patch and the most effective one i have ever used
and it didn't come out from nothing
don't you know the famous 200 lines that improves linux ?
http://www.webupd8.org/2010/11/alter...nel-patch.html
http://askubuntu.com/questions/13562...200-line-patch
http://askubuntu.com/questions/19051...x-kernel-patch
http://linux.slashdot.org/story/10/1...t-Does-Wonders

i think it does do something after all

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

Not using them so cannot comment.

Anyway, a little progress, notifications semi-working, at last I grok the damn swap remap thing, hopefully by 1-2 days will have them fully working. And will reflash my second device so comparison with/without compcache to be on a common base.

TBH right now i am using stock settings for swapiness & co, only overclock to 900. With compcache(and partial notifications) enabled the damn thing is almost like a butter with ovi maps + facebook + youtube. Of course once we hit internal flash lag starts.

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

And you were told hundreds of times that it is already implemented by nokia so "magic patch" does nothing on cgroups side. God knows what exactly it does. Anyway lets stay on topic please.

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

the best thing is to stick on the topic please
and freemangordon that's what i ment before
when reaching the internel swap N900 starts to lag till have to reboot

just an extra info
cgrouping in /dev gives a better performance than cgrouping in /sysparts

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

I re-flashed my second N900 (rootfs & eMMC) and then installed just rootsh + Compcache kernel-power.
It seems to work fine. Some lagging when ramzswap is full, but still responding quite OK.

Other N900, (KP+CSSU+speedpatch, etc...) started to lag heavily after ramzswap was full and eventually needed a reboot (or battery removal). I tried with speedpatch installed / removed, results was the same. Something is confiliction with compcache.

I'll add CSSU and other stuff to clean N900 and check what will happen.

EDIT: Installation steps (cumulative)
1) CSSU installed -> OK, no changes
2) Swappolube installed and "proposed values" applied -> NOK (stop responding, battery removal needed)
3) Swappolube and just swappiness adjusted to 30 -> OK (better that proposed values, but lots of lagging when heavily loaded)
- Annoying lagginess starts somewhere when +50M flash swap is in use. Below that it is quite usable.
4) Speedpatch (speedpatch_0.4.deb) -> Best performance so far. Ramzswap full & +70M flash swap and device is very usable and lagging is very minimal ( I had to test this even agreed to keep it in separate topic)

EDIT_2:Copied same settings for the other N900.

EDIT_3:With the other N900 problems started again after some usage. Load averages are showing: ~10, ~8 and ~3, but there is no program using CPU more that 2% and CPU freq is at min (min = 500MHz, max=805MHz). This ghost loading is causing lack of responsiveness. Any ideas how to find out what is causing extremely high constant loading? Top and Conky are not giving any additional usable information.

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

...it is? Wasn't the kernel auto-scheduler patch first upstream included in kernel 2.6.3x?

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

@Gusse
you're using the suggested disksize?

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

Yep. disksize_kb=131073

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

Which is expected behaviour, once out of RAM (and notifications turned off, so memory consumption is suboptimal), we hit internal flash swap. The benefit is that we can run more applications before that hit. Lets see how it will behave once notifications are running

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

So karam's patch adds upstream kernel patch to n900 kernel? Don't think so.

Again, stay on topic please. I don't want to further discuss speedpatch here, there is a dedicated thread for that. Thanks.

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

This gives me an epic boner.

Just sayin'

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

Oh I swear to $DEITY that we should ban everyone that advocates this 200-line patch.
Nokia tries to do this and fails - see ohmd/syspart.
The downside to Nokia's approach is that it's too much on top of the kernel's SLUB. Nokia should've focused on making "nice" actually work. Most programs managed by syspart are small-memory-footprint programs that run in the background. Except for browserd and tracker (but that should be nice'd to 20)

Unfortunately, for some reason the device becomes unusable without syspart/ohmd.

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

OH gosh Hurrian I swear to GOD that my device is usable
and pleeeeease stay on topic

i don't think that is necessary -_-

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

Then go post about your "magic" patch in the respectable thread.

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

Here we go, first post updated (swap notifications working most of the time)

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

http://talk.maemo.org/showpost.php?p...&postcount=979

Go away yourself!

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

will this be part of KP49?

and If you install the deb (which the link to is dead BTW), do you still need to copy the *.ko file?

thanks

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

freemangordon, any chance you can upload the debs to goput.it? Seems that I need to register to use abv.bg.

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

Sorry about that, got the impression that registration is not needed for download.

http://goput.it/n0x.zip

EDIT: link should work now.

Re: [ANNOUNCE] COMPCACHE on kernel-power
 

Who is there? The guy with the swap performance measurements? Great, would you measure compcache swap performance?

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

You're the man, freemangordon!
Running compcache with 128MB disksize, xterm, microb, contacts, conversations, fapman, and LibreOffice running. It doesn't explode anymore!

Some things get laggy under high load though (such as CSSU's tactile) and the UI. Music playback is OK.

BTW, my 512MB swap partition which used to reach 160 MB under this load now contains a bit more than 7MB. Magical!

EDIT: Unfortunately, while stable enough for me, it sends CPU usage to high hell. 100% most of the time, running at 250-600MHz btw.

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

High CPU usage seems to be because of the disksize you are using, try with 96 MB.

And now as we have compcache running I am sure someone have to play a little with kernel parameters (swappiness, page size, what else?) in order to have maximum performance gain

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

LibreOffice running? How?

Can you give me the steps to install it?

Thanks

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

Install Debian and have your LibreOffice running in chroot :)

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

have you got speedpatch installed?
sorry my off topic

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

^ *Kills this one with a railgun*

sorry for off-topic even more, but LibreOffice *don't* work well via EasyDebian, due to gconf2 upgrade, which break almost everything in ED...

@freemangordon, what do you mean by "notifications working *most of the time*"?

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

Strange, works OK with disksize 131072 but with disksize 120512 I get instant reboot every time.

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

Damn, this is exciting!

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

@estel:
from first post:
"Notifications are working, but seems there are paths in the kernel swap subsystem allowing direct write over a page without first marking it free.
Seems like memory consumption should not be affected by that.
"
aanckar: are you sure you use flash swap too? if not you will have oom kills for sure

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

May I have your /dev/mtd2 contents?

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

False alarm, seems only to have been caused by a bit noobishness from my part (an error when typing the swapon/swapoffs), both work well now.

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

do you need a SD card to use this?

Re: [ANNOUNCE] COMPCACHE on kernel-power (now with notifications)
 

To all of you who are trying compcache:

Please use disksize_kb of 120512, it seems 131072 leads to massive lag after a while(not enough free RAM left).

EDIT:
and please test with different values for /proc/sys/vm/page-cluster and /proc/sys/vm/swappiness.

My current values are
swappiness 20
page-cluster 2

but my flash swap is on SD card, so for internal flash optimal values may vary