Jens Axboe recently posted a patch for the 2.4 stable kernel introducing a "laptop_mode" sysctl variable aimed at increasing battery time on laptops. The patch accomplishes this in a couple of main ways. First, when a specific write happens, the systems finds any other data that is also ready to be written and writes it at the same time. Additionally, whenever a read happens, the system again finds any data that is ready to be written and writes it following the read. In each case, the idea is to minimize the frequency that a laptop's hard drive has to spin up, thus prolonging battery life.

When posting a second version of the patch with "a few tweaks and adjustments", Jens noted, "I've gotten several mails [from people] who really like the patch and [say] that it really adds a [non-insignificant] amount of extra battery time. I consider the patch final at this point." The patch is against the 2.4.21-rc2 stable kernel, though Jens acknowledged that he intends to port the patch to the 2.5 development kernel in the near future. Read on for much of the resulting discussion.

From: Jens Axboe
To: Linux Kernel 
Subject: [PATCH] 2.4 laptop mode
Date: Wed, 14 May 2003 11:35:04 +0200

Hi,

Now, this isn't the prettiest patch in the world. But it does allow me
to get good spin down times on my laptop hard drive. It was somewhat
inspired by the 2.5.early version akpm did. Basically, it adds:

- a laptop_mode sysctl variable that controls:

	- write back of dirty data. when we have to do it, write back
	  all of it.

	- when a read is scheduled, write out dirty pending data right
	  after. purpose of that is to make good use of a disk that was
	  potentially spun down before.

- a block_dump sysctl variable. when this is set, it dumps info to dmesg
  about which process scheduled the read/write. this is helpful for
  determining _why_ the disk is spinning up all the time.

Patch also sets expire time of buffer and kupdated intervals to 10
minutes, this should probably just be done from a script. ext3's jbd is
changed to set the journal expire time to the same as the buffer expire
time.

[patch]
-- 
Jens Axboe



From: Martin Waitz
Subject: Re: [PATCH] 2.4 laptop mode
Date: Thu, 15 May 2003 10:59:12 +0200

hi :)

On Wed, May 14, 2003 at 11:35:04AM +0200, Jens Axboe wrote:
> Now, this isn't the prettiest patch in the world. But it does allow me
> to get good spin down times on my laptop hard drive. It was somewhat
> inspired by the 2.5.early version akpm did. Basically, it adds:

if you are interested in spinning down hard drives, you might want to read
http://www4.informatik.uni-erlangen.de/Publications/pdf/
Weissel-Beutel-Bellosa-OSDI-CoopIO.pdf

they describe strategies to get maximum sleep times for drives by
bundling accesses to hard discs.

they even go a little bit faster and allow user space to give hints
about when they need data.
i only had a brief look at the sources but i guess this could be folded
into the aio interface.
(CoopIO as described above adds its own system calls)
it would be great if somethink like that could be ported to 2.5...

-- 
CU,		  / Friedrich-Alexander University Erlangen, Germany
Martin Waitz	//  Department of Computer Science 3       _________
______________/// - - - - - - - - - - - - - - - - - - - - ///
dies ist eine manuell generierte mail, sie beinhaltet    //
tippfehler und ist auch ohne grossbuchstaben gueltig.   /



From: Jens Axboe
Subject: Re: [PATCH] 2.4 laptop mode
Date: Thu, 15 May 2003 12:06:53 +0200

On Thu, May 15 2003, Martin Waitz wrote:
> hi :)
> 
> On Wed, May 14, 2003 at 11:35:04AM +0200, Jens Axboe wrote:
> > Now, this isn't the prettiest patch in the world. But it does allow me
> > to get good spin down times on my laptop hard drive. It was somewhat
> > inspired by the 2.5.early version akpm did. Basically, it adds:
> 
> if you are interested in spinning down hard drives, you might want to read
> http://www4.informatik.uni-erlangen.de/Publications/pdf/
> Weissel-Beutel-Bellosa-OSDI-CoopIO.pdf
> 

Interesting, I did not know about this paper. A lot of what my patch
does is identical to what they describe, I stop at the OS level though
and haven't (and don't really want to) extend it to applications. I
think that we can get 'pretty good' power saves without going that extra
mile.

Note that I also operate outside of the 4 ide power states described,
sleep -> standby -> idle -> active. I chose to disregard sleep, because
it requires a reset and drive program when transitioning from sleep to
idle. It's my feeling that most drives do the idle -> active transition
(and vice versa) on their own. So for me, that just leaves on
interesting operating mode (idle-active :). However, I added the
acoustic management in-between that. So when the drive is considered
'active' (ie serving requests), the amount of io will determine how fast
the seeks go by switching between the acoustic levels.

So with my patch, we are pretty close to the ECU level described in the
paper. With the laptop patch, we handle the writeout of dirty data at
appropriate times (when reads spin up the disk, etc) as well.

> they describe strategies to get maximum sleep times for drives by
> bundling accesses to hard discs.

I bundle writes with reads (slighly postponed), doing more would require
the added new syscalls.

> they even go a little bit faster and allow user space to give hints
> about when they need data.
> i only had a brief look at the sources but i guess this could be folded
> into the aio interface.
> (CoopIO as described above adds its own system calls)

Yeah, using aio would make it a lot easier and wouldn't require many
changes to the existing aio interface.

> it would be great if somethink like that could be ported to 2.5...

What's stopping you?

-- 
Jens Axboe



From: Martin Waitz
Subject: Re: [PATCH] 2.4 laptop mode
Date: Thu, 15 May 2003 14:54:13 +0200

hi :)

On Thu, May 15, 2003 at 12:06:53PM +0200, Jens Axboe wrote:
> > it would be great if somethink like that could be ported to 2.5...
> What's stopping you?

well the group that did that work does not have enough money to
fund new staff, so i can't get paid to do it...

and i don't have enough free time :(

but anyway, your work on this topic is great, too! :)
i am really looking forward to having a laptop-optimized os

-- 
CU,		  / Friedrich-Alexander University Erlangen, Germany
Martin Waitz	//  Department of Computer Science 3       _________
______________/// - - - - - - - - - - - - - - - - - - - - ///
dies ist eine manuell generierte mail, sie beinhaltet    //
tippfehler und ist auch ohne grossbuchstaben gueltig.   /



From: Jens Axboe
Subject: Re: [PATCH] 2.4 laptop mode
Date: Thu, 15 May 2003 17:14:28 +0200

On Thu, May 15 2003, Martin Waitz wrote:
> hi :)
> 
> On Thu, May 15, 2003 at 12:06:53PM +0200, Jens Axboe wrote:
> > > it would be great if somethink like that could be ported to 2.5...
> > What's stopping you?
> 
> well the group that did that work does not have enough money to
> fund new staff, so i can't get paid to do it...
> 
> and i don't have enough free time :(

Ah

> but anyway, your work on this topic is great, too! :)
> i am really looking forward to having a laptop-optimized os

Thanks, I'll surely move it to 2.5 at some point in the near future. But
only to the extent of what it does know, I don't care enough to take it
further I'm afraid. The benefit would not justify the time spent,
compared to what the OS only solution gets me :)

-- 
Jens Axboe

From: Jens Axboe
Subject: [PATCH] laptop mode, #2
Date: Fri, 16 May 2003 13:33:09 +0200

Hi,

Made a few tweaks and adjustments:

- If block_dump is set, also dump who is marking a page/buffer as dirty.
  akpm recommended this.

- Don't touch default bdflush parameters (see script)

That's about it. I've gotten several mails who really like the patch and
that it really adds a non-significant amount of extra battery time. I
consider the patch final at this point.

Patch is against 2.4.21-rc2 (ish)

[patch]

[control script]

-- 
Jens Axboe



From: Pavel Machek
Subject: Re: [PATCH] laptop mode, #2
Date: Fri, 16 May 2003 21:55:36 +0200

Hi!

> Made a few tweaks and adjustments:
> 
> - If block_dump is set, also dump who is marking a page/buffer as dirty.
>   akpm recommended this.
> 
> - Don't touch default bdflush parameters (see script)
> 
> That's about it. I've gotten several mails who really like the patch and
> that it really adds a non-significant amount of extra battery
> time. I

Non-significant? Like it adds no time at all?

> consider the patch final at this point.
> 
> Patch is against 2.4.21-rc2 (ish)

It looks nice, but without Documentation/laptop_mode, I can't say much
more ;-)))).
								Pavel
-- 
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]



From: Jens Axboe
Subject: Re: [PATCH] laptop mode, #2
Date: Sat, 17 May 2003 10:16:58 +0200

On Fri, May 16 2003, Pavel Machek wrote:
> Hi!
> 
> > Made a few tweaks and adjustments:
> > 
> > - If block_dump is set, also dump who is marking a page/buffer as dirty.
> >   akpm recommended this.
> > 
> > - Don't touch default bdflush parameters (see script)
> > 
> > That's about it. I've gotten several mails who really like the patch and
> > that it really adds a non-significant amount of extra battery
> > time. I
> 
> Non-significant? Like it adds no time at all?

Woops, that was reversed :)

> > consider the patch final at this point.
> > 
> > Patch is against 2.4.21-rc2 (ish)
> 
> It looks nice, but without Documentation/laptop_mode, I can't say much
> more ;-)))).

The attached script basically explained everything, but yes I will add a
laptop_mode file as well.

-- 
Jens Axboe

• Google archive of above thread
• Patch
• Control Script