| 1 |
1275 |
phoenix |
Laptop mode
|
| 2 |
|
|
===========
|
| 3 |
|
|
|
| 4 |
|
|
This small doc describes the 2.4 laptop mode patch.
|
| 5 |
|
|
|
| 6 |
|
|
Last updated 2003-05-25, Jens Axboe
|
| 7 |
|
|
|
| 8 |
|
|
Introduction
|
| 9 |
|
|
------------
|
| 10 |
|
|
|
| 11 |
|
|
A few properties of the Linux vm makes it virtually impossible to attempt
|
| 12 |
|
|
to spin down the hard drive in a laptop for a longer period of time (more
|
| 13 |
|
|
than a handful of seconds). This means you are lucky if you can even reach
|
| 14 |
|
|
the break even point with regards to power consumption, let alone expect any
|
| 15 |
|
|
decrease.
|
| 16 |
|
|
|
| 17 |
|
|
One problem is the age time of dirty buffers. Linux uses 30 seconds per
|
| 18 |
|
|
default, so if you dirty any data then flusing of that data will commence
|
| 19 |
|
|
at most 30 seconds from then. Another is the journal commit interval of
|
| 20 |
|
|
journalled file systems such as ext3, which is 5 seconds on a stock kernel.
|
| 21 |
|
|
Both of these are tweakable either from proc/sysctl or as mount options
|
| 22 |
|
|
though, and thus partly solvable from user space.
|
| 23 |
|
|
|
| 24 |
|
|
The kernel update daemon (kupdated) also runs at specific intervals, flushing
|
| 25 |
|
|
old dirty data out. Default is every 5 seconds, this too can be tweaked
|
| 26 |
|
|
from sysctl.
|
| 27 |
|
|
|
| 28 |
|
|
So what does the laptop mode patch do? It attempts to fully utilize the
|
| 29 |
|
|
hard drive once it has been spun up, flushing the old dirty data out to
|
| 30 |
|
|
disk. Instead of flushing just the expired data, it will clean everything.
|
| 31 |
|
|
When a read causes the disk to spin up, we kick off this flushing after
|
| 32 |
|
|
a few seconds. This means that once the disk spins down again, everything
|
| 33 |
|
|
is up to date. That allows longer dirty data and journal expire times.
|
| 34 |
|
|
|
| 35 |
|
|
It follows that you have to set long expire times to get long spin downs.
|
| 36 |
|
|
This means you could potentially loose 10 minutes worth of data, if you
|
| 37 |
|
|
set a 10 minute expire count instead of just 30 seconds worth. The biggest
|
| 38 |
|
|
risk here is undoubtedly running out of battery.
|
| 39 |
|
|
|
| 40 |
|
|
Settings
|
| 41 |
|
|
--------
|
| 42 |
|
|
|
| 43 |
|
|
The main knob is /proc/sys/vm/laptop_mode. Setting that to 1 switches the
|
| 44 |
|
|
vm (and block layer) to laptop mode. Leaving it to 0 makes the kernel work
|
| 45 |
|
|
like before. When in laptop mode, you also want to extend the intervals
|
| 46 |
|
|
desribed above. See the laptop-mode.sh script for how to do that.
|
| 47 |
|
|
|
| 48 |
|
|
It can happen that the disk still keeps spinning up and you don't quite
|
| 49 |
|
|
know why or what causes it. The laptop mode patch has a little helper for
|
| 50 |
|
|
that as well, /proc/sys/vm/block_dump. When set to 1, it will dump info to
|
| 51 |
|
|
the kernel message buffer about what process caused the io. Be very careful
|
| 52 |
|
|
when playing with this setting, it is advisable to shut down syslog first!
|
| 53 |
|
|
|
| 54 |
|
|
Result
|
| 55 |
|
|
------
|
| 56 |
|
|
|
| 57 |
|
|
Using the laptop-mode.sh script with its default settings, I get the full
|
| 58 |
|
|
10 minutes worth of drive spin down. Provided your work load is cached,
|
| 59 |
|
|
the disk will only spin up every 10 minutes (well actually, 9 minutes and 55
|
| 60 |
|
|
seconds due to the 5 second delay in flushing dirty data after the last read
|
| 61 |
|
|
completes). I can't tell you exactly how much extra battery life you will
|
| 62 |
|
|
gain in laptop mode, it will vary greatly on the laptop and workload in
|
| 63 |
|
|
question. The only way to know for sure is to try it out. Getting 10% extra
|
| 64 |
|
|
battery life is not unrealistic.
|
| 65 |
|
|
|
| 66 |
|
|
Notes
|
| 67 |
|
|
-----
|
| 68 |
|
|
|
| 69 |
|
|
Patch only changes journal expire time for ext3. reiserfs uses a hardwire
|
| 70 |
|
|
value, should be trivial to adapt though (basically just make it call
|
| 71 |
|
|
get_buffer_flushtime() and uses that). I have not looked at other
|
| 72 |
|
|
journalling file systems, I'll happily accept patches to rectify that!
|
