56 replies to this topic

[MiLo]

maybe it helps if zram has more priority and work as is should.

Normally, zram will be your one-and-only device, so priority won't matter.
 

swapon -p 100 /dev/zram0

Try this on your box, and notice that busybox's swapon does not support the -p option.
 

and they talk about using cleancache&frontswap to activate for kernel 3.10.

Interesting read. Unfortunately, we only have one single box with a kernel >= 3.10, the rest is far behind that, and without the manufacturer's support, we can't move on. Otherwise, we'd have had 3.12 kernels on all boxes before Xmas...

[MiLo]

"If you have kernl 3.10 I would suggest not using zram for swap as there are better alternatives.  More later.

Our boxes have relatively slow CPUs, and usually no disk swap storage. So likely even on the newer kernels, zram will perform better on this particular set of hardware/software than zcache and friends.

Our "disk pages" are usually filled with compressed audio/video data, and these are totally uncompressible, so any attempt to compress these pages are a waste of those precious CPU cycles we have. So that rules out "cleancache" and "frontswap" from doing anything useful - no point attempting to compress disk data, it will be uncompressible.

The memory that we have that is compressible are the pages filled with program code and (non-video/audio) program data (e.g. bitmaps and EPG data). Which implies that you should set "swappiness" to its default or higher. The kernel can move these to a swapfile like zram if memory runs tight. The idea behind zram was to have some "spare" things to do when memory is running low, and to move hardly used pages to a compressed area of memory. In particular, I hoped that the EPG data could be moved there...

[Erik Slagter]

I was hoping adding some sort of swap (either "classic" or zram) would allow the kernel to swap out pages, swap them in again, and in that proces create more contiguous ranges of (physical) memory, by swapping them in at other pages' addresses.

[MiLo]

If the kernel would do that, it could also "shuffle" the pages and data around to make a region continuous. Which it doesn't. I guess. I never looked into this part in great detail though. But theoretically, it's quite possible to 'make' a contiguous memory region available by just copying its contents elsewhere, adjust the memory table, and use the now free contiguous range for the request.

(I was wondering if zcache and friends could make use of an embedded FPGA to compress pages and store them outside the chip somewhere... although usually there's more than enough RAM available on SOCs, but it makes for a nice friday afternoon experiment...)

[ims]

tried use zram on xp1000 :

root@xp1000:~#opkg install zram

...  downloaded and instaled

 Removing any system startup links for zram.sh ...

Configuring kernel-module-zram.

Configuring zram.

 Adding system startup for /etc/init.d/zram.sh.

/etc/rc3.d/S05zram.sh: line 53: can't create /sys/block/zram*/reset: nonexistent directory

root@xp1000:~# mkswap /dev/zram0

mkswap: image is too small

root@xp1000:~# swapon -a 100 /dev/zram0

/dev/+zram0 is created now. My subjective feeling is, that E2 works faster. May be, better performance made all new patches for XP1000 or zram ?

But why this messages ?

[Taapat]

@ims if you entered only this commands, the my subjective feeling is when you're not using swap in zram.

Update kernel to load the zram as a module with the zram.sh script.

If you are setup manually, then specify the zram disk size to avoid error mkswap: image is too small.

See an example in init script /etc/init.d/zram.sh.

[Erik Slagter]

If the kernel would do that, it could also "shuffle" the pages and data around to make a region continuous. Which it doesn't. I guess. I never looked into this part in great detail though. But theoretically, it's quite possible to 'make' a contiguous memory region available by just copying its contents elsewhere, adjust the memory table, and use the now free contiguous range for the request.

AFAIK the kernel doesn't move pages around, I think because it's not necessary when all drivers are behaving themselves  I don't think the kernel remembers where a page comes from, I can't think of reason for that, so that would mean it would get swapped back to a "random" position. Then it's the question how smart the kernel is with swapping back, if it goes out of it's way searching for the smallest holes, or just picks the "first" free. I guess pieterg can give the authoritative answer

[ims]

@ims if you entered only this commands, the my subjective feeling is when you're not using swap in zram.

Update kernel to load the zram as a module with the zram.sh script.

If you are setup manually, then specify the zram disk size to avoid error mkswap: image is too small.

See an example in init script /etc/init.d/zram.sh.

Yes, only install zram is enough. As I saw, for my XP1000 is created 117908kB. But it is still free. It is not used yet or meminfo did not display it ? Or must I some set ? Here is meminfo on picture. SwapTotal = 117908kB, SwapFree = 117908kB

Attached Files

•  memory_xp1000.jpg   83.02KB   34 downloads

[ims]

when enigma starts on XP1000 , there is message: 

couldn't read model:  [Errno 2] No such file or directory: '/sys/block/zram0/device/model'

Can it be ? 

Edited by ims, 13 January 2014 - 21:49.

[Taapat]

Can it be ? 

 

Pay attention to this:

Update kernel to load the zram as a module with the zram.sh script.

 

For example, see here: http://openpli.org/f...uo/#entry398745

Edited by Taapat, 14 January 2014 - 08:10.

[ims]

i am using last clear build image...

[Taapat]

Sorry, I will not read the error message carefully.
I think it comes from Harddisk.py which confusing mount points.

[MiLo]

Explain.

[Taapat]

I have seen this error in other logs after enabling zram. The only place where I've found in enigma where is looking for model and has following error is: http://sourceforge.n...arddisk.py#l770
I do not have mipsel receiver, and even though I am using zram, I do not have that errors, and therefore I have not looked anything more.

[Taapat]

I think that the need to add zram to blacklisted: http://sourceforge.n...arddisk.py#l634

[MiLo]

There's no problem there. The zram device is a block device like any other.

You can also format a zram block device in ext4 and use it as a harddisk. Blacklisting it would be wrong for that reason. It is a block device, and it can be mounted like any other block device.

I fail to see what the problem is you're trying to solve. If you use zram for swap, it doesn't show up in the media list, for example.

[WanWizard]

I do not have mipsel receiver, and even though I am using zram, I do not have that errors, and therefore I have not looked anything more.

 

suggests this is not about an OpenPLi image?