Compcache

Compcache or Ramzswap, hitherto referred as ramz is a method for taking a chunk of 'normal' RAM and using it as a compressed virtual memory (VM) swap space. That is, we use RAM as a sort of a virtual, virtual RAM by compressing the contents on the fly and storing it in RAM that's marked up as a swap device. This is in order to squeeze as much space from the limited RAM as possible. A swap device created from compressed RAM is many times faster than a regular swap device using hard drives or traditional flash memory.

N8x0

To be added by a wild N8x0 user...
Apparently compcache is the single greatest thing to ever happen to the N8x0.

N900

The problem

---

Lack of RAM is considered one on the main performance bottlenecks a N900 user has to contend with. In its stock set-up, the N900 is prone to having massive swapping after even short periods of heavy use. This leads to poor performance, and in the worst cases, a reboot. This is generally alleviated in one or several of four ways.

The solutions

---

Install 'Swappolube'

Swappolube (sic) is a set of VM parameter tweaks initially dreamt up by the user Hawaii. For MOST users, the installation of Swappolube from the repositories and applying the default settings will suffice to alleviate the problem.

Swap on micro SD card(uSD)

It has been found that an easy way to alleviate eMMC contention is to move the swap space to the uSD card[1].

Swap de-fragmentation

It has been found that as the swap space fills up, the OS is eventually more and more unable to allocate continuous space on the swap device. This leads to poor performance, that can be periodically reset by turning the swapping device off and on again, clearing the slate. The package ereswap was created to make this easy.

Compcache

The article's namesake. The use of compcache has been a contentious issue. Most users have found it to cause extremely poor system performance. For many users, as soon as the ramz space gets full - which is pretty quick, considering it's only tens of megabytes large - the system grinds to a slow crawl.

Some empirical and anecdotal evidence suggests that the 'extreme' swap-retention settings of swappolube (that most users run) are highly unsuitable ramz. Swappolube sets the VM to swap as rarely as possible. This does not work for ramz - likely, an opposite approach must be taken: The device must swap a lot. As in this scenario the swapping is optimally mostly done to the RAM swap device, there will be little penalty in speed.

Using compcache

---

Caveats!

• While ramz does work on kernel power version 49 (KP49), it is highly recommended that you use KP50!

• Ramz requires a disksize that is specified in KB in powers of 2. Acceptable sizes are:

Ramz size	English
2048	2MB
4096	4MB
8192	8MB
16384	16MB
32768	32MB
65536	64MB
98304	96MB
131072	128MB

• It is recommended that you do not use more than 64MB for ramz.

• Unless enabled very early in the device boot, you MUST have swap enabled on a physical disk before enabling ramz, or your n900 WILL insta-crash and reboot, potentially corrupting your file system!

• Due to the CPU load slightly increased by the de/compression done by ramz, using ramz likely incurs a slight penalty in battery life.

• Compcache is still considered experimental! It may not work, asplode your n900 and rip your *@!# clean off. You have been warned.

How to enable compcache

To enable compcache you must enter these commands as root, the following will create a ramz disk of 64MB:

<source lang="bash"> insmod /lib/modules/current/ramzswap.ko disksize_kb=65536 </source>

It will take a short while for the kernel to catch up with this, so you should pair this with a 'sleep 1' command. Once enabled, you must tell the system to use the ramz disk:

<source lang="bash"> swapon /dev/ramzswap0 </source>

For any of this to be any benefit, the ramz disk needs to have a higher priority than the 'physical' swap space. Make sure ALL applications are closed and disable then re-enable your physical swap. On a stock setup (swap on eMMC) you can do this:

<source lang="bash"> swapoff /dev/mmcblk0p3 && sleep 1 && swapon /dev/mmcblk0p3 </source>

For other setups, i.e. when you have swap on uSD you must substitute 'mmcblk0p3' with the location of your swap space. (If your swap partition is the second partition on the uSD then it will be 'mmcblk1p2').

You can now check you have the appropriate partitions with:

<source lang="bash"> cat /proc/swaps Filename Type Size Used Priority /dev/mmcblk0p3 partition 786424 64476 -2 /dev/ramzswap0 partition 65528 61572 -1 </source>

You can see ramzswap is 64MB AND with a higher priority than mmcblk0p3.

Now you must change your VM settings to effectively use the new ramz setup. The following settings have been found to work with 32MB, 64MB and 128MB (beware!) ramz sizes. They are a work-in-progress and will be updated as they are tuned to the authors' workload (heavy user, treating n900 as a pocket laptop as oppose d to merely a phone).

Paste the below lines into xterm to setup the VM and tune the flash device usage:

<source lang="bash"> echo 10 > /proc/sys/vm/dirty_ratio echo 10 > /proc/sys/vm/dirty_background_ratio echo 250 > /proc/sys/vm/dirty_writeback_centisecs echo 1000 > /proc/sys/vm/dirty_expire_centisecs echo 100 > /proc/sys/vm/swappiness echo 200 > /proc/sys/vm/vfs_cache_pressure echo 0 > /proc/sys/vm/page-cluster echo 64 > /sys/block/mmcblk0/queue/nr_requests echo 64 > /sys/block/mmcblk1/queue/nr_requests echo 0 > /sys/block/mmcblk0/queue/iosched/slice_idle echo 0 > /sys/block/mmcblk1/queue/iosched/slice_idle echo 32 > /sys/block/mmcblk0/queue/iosched/quantum echo 32 > /sys/block/mmcblk1/queue/iosched/quantum echo 100 > /sys/block/mmcblk0/queue/iosched/fifo_expire_sync echo 100 > /sys/block/mmcblk1/queue/iosched/fifo_expire_sync echo 200 > /sys/block/mmcblk0/queue/iosched/fifo_expire_async echo 200 > /sys/block/mmcblk1/queue/iosched/fifo_expire_async </source>

How to do this all at once

---

Paste the below code into a root xterm.

<source lang="bash">

1. !/bin/sh

1. referance: http://www.westnet.com/~gsmith/content/linux-pdflush.htm

1. Smart swap on/off
2. drivelist=$(sfdisk -lnd /dev/mmcblk1)

if [ "$(cat /proc/swaps | awk '/mmcblk/ {print $1}' | cut -d "k" -f2 | cut -c1)" == "0" ]; then

       echo "swap is on emmc"
       if [ "$(sfdisk -lnd /dev/mmcblk0 | grep Id=82 | awk '/mmcblk0/ {print $1}')" ]; then
       nice -20 swapoff $(sfdisk -lnd /dev/mmcblk0 | grep Id=82 | awk '/mmcblk0/ {print $1}')
       #insert ramzswap module and specify disksize
       insmod /lib/modules/current/ramzswap.ko disksize_kb=65536 && sleep 1
       nice -20 swapon $(sfdisk -lnd /dev/mmcblk0 | grep Id=82 | awk '/mmcblk0/ {print $1}')
       echo "swap is on swapped on emmc"
       fi

fi

if [ $(cat /proc/swaps | awk '/mmcblk/ {print $1}' | cut -d "k" -f2 | cut -c1) = "1" ]; then

       nice -20 swapoff $(sfdisk -lnd /dev/mmcblk1 | grep Id=82 | awk '/mmcblk1/ {print $1}')
       #insert ramzswap module and specify disksize
       insmod /lib/modules/current/ramzswap.ko disksize_kb=65536 && sleep 1
       nice -20 swapon $(sfdisk -lnd /dev/mmcblk1 | grep Id=82 | awk '/mmcblk1/ {print $1}')
       echo "swap is on sd"

fi

1. vm changes
2. /proc/sys/vm/dirty_ratio (default 40)

echo 10 > /proc/sys/vm/dirty_ratio

1. dirty_background_ratio

echo 10 > /proc/sys/vm/dirty_background_ratio

1. dirty_writeback_centisecs

echo 250 > /proc/sys/vm/dirty_writeback_centisecs

1. /proc/sys/vm/dirty_expire_centiseconds (default 3000)

echo 1000 > /proc/sys/vm/dirty_expire_centisecs

1. free space saved for large allocations, values below 256 cause insta-crash
2. echo 1024 > /proc/sys/vm/min_free_kbytes

1. 'Normal' VM stuff

echo 100 > /proc/sys/vm/swappiness echo 200 > /proc/sys/vm/vfs_cache_pressure echo 0 > /proc/sys/vm/page-cluster

1. Flash disc tuning

echo 32 > /sys/block/mmcblk0/queue/nr_requests echo 32 > /sys/block/mmcblk1/queue/nr_requests echo 0 > /sys/block/mmcblk0/queue/iosched/slice_idle echo 0 > /sys/block/mmcblk1/queue/iosched/slice_idle echo 32 > /sys/block/mmcblk0/queue/iosched/quantum echo 32 > /sys/block/mmcblk1/queue/iosched/quantum echo 100 > /sys/block/mmcblk0/queue/iosched/fifo_expire_sync echo 100 > /sys/block/mmcblk1/queue/iosched/fifo_expire_sync echo 200 > /sys/block/mmcblk0/queue/iosched/fifo_expire_async echo 200 > /sys/block/mmcblk1/queue/iosched/fifo_expire_async

1. display results

cat /proc/swaps </source>

Automating all of this

---

There is a package called swapset[2], which attempts to setup swaps and the ramz module automatically. The author of this article has not used it, and thus cannot vouch for its effectiveness. Using swapset AND putting the above settings into swappolube would probably cause all of the above to work automatically. Probably.

N9/50

To be added by an ardent masochist.