From chm.duquesne at gmail.com Wed Feb 1 09:34:45 2012 From: chm.duquesne at gmail.com (Christophe-Marie Duquesne) Date: Wed, 1 Feb 2012 09:34:45 +0100 Subject: [laptop-mode] laptop-mode-tools loops on lock_retry() In-Reply-To: <4F28E645.3050603@researchut.com> References: <4F2847B1.3050805@researchut.com> <4F28E645.3050603@researchut.com> Message-ID: Thank you! This is indeed fixing my problems. Now LMT correctly modifies lcd brightness, takes care of the usb, and all the other things it was supposed to do. Cheers, Christophe-Marie From laptop-mode at yungchin.nl Sat Feb 18 01:08:44 2012 From: laptop-mode at yungchin.nl (yungchin) Date: Sat, 18 Feb 2012 00:08:44 +0000 Subject: [laptop-mode] laptop-mode with btrfs Message-ID: Dear list, Does anyone here have experience running laptop-mode with a btrfs filesystem? I seem to be stuck with btrfs-submit-1 and btrfs-transacti waking the disk up all the time (lm-profiler reports around 100 write operations for these two processes over 600 seconds, on an idle system, on battery). I'm trying to find out what these writes are for, but I'm not finding much of interest, eg: yungchin at D630new:~$ sudo btrfs subvolume find-new / 7810 transid marker was 7811 (usually this would list files that had changed, I think the empty list means no file data were written in this transaction, only journal or metadata - yet it spun up the disk for this...) This is a laptop running what is mostly debian squeeze, but with a squeeze-backports 3.2 kernel and laptop-mode-tools 1.60 from sid. Btrfs is the root FS, and (just to make life complicated I suppose) it sits on a LUKS-encrypted LVM volume. I've also experimented with the nobarrier mount option (which avoids flushing the disk hardware cache, which I have disabled anyway) but as you might probably expect that makes no difference. I'd be grateful for any input! Thanks, Yung-Chin From rrs at researchut.com Sat Feb 18 07:30:41 2012 From: rrs at researchut.com (Ritesh Raj Sarraf) Date: Sat, 18 Feb 2012 12:00:41 +0530 Subject: [laptop-mode] laptop-mode with btrfs In-Reply-To: References: Message-ID: <4F3F4591.7060602@researchut.com> This is an OpenPGP/MIME signed message (RFC 2440 and 3156) --------------enig154741EB0404356D944F9B0E Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable On Saturday 18 February 2012 05:38 AM, yungchin wrote: > Dear list, > > Does anyone here have experience running laptop-mode with a btrfs files= ystem? > > I seem to be stuck with btrfs-submit-1 and btrfs-transacti waking the > disk up all the time (lm-profiler reports around 100 write operations > for these two processes over 600 seconds, on an idle system, on > battery). I'm trying to find out what these writes are for, but I'm > not finding much of interest, eg: > > yungchin at D630new:~$ sudo btrfs subvolume find-new / 7810 > transid marker was 7811 > > (usually this would list files that had changed, I think the empty > list means no file data were written in this transaction, only journal > or metadata - yet it spun up the disk for this...) > > Recently I was viewing the fosdem video and the very same question was asked there. It was told that the file system writes some metadata every few seconds (IIRC, it was mentioned 30 seconds) but that that was tunable= =2E --=20 Ritesh Raj Sarraf RESEARCHUT - http://www.researchut.com "Necessity is the mother of invention." --------------enig154741EB0404356D944F9B0E Content-Description: OpenPGP digital signature Content-Disposition: attachment; filename="signature.asc" MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 900 bytes Desc: OpenPGP digital signature URL: --------------enig154741EB0404356D944F9B0E-- From laptop-mode at yungchin.nl Sat Feb 18 23:37:23 2012 From: laptop-mode at yungchin.nl (yungchin) Date: Sat, 18 Feb 2012 22:37:23 +0000 Subject: [laptop-mode] laptop-mode with btrfs In-Reply-To: References: <4F3F4591.7060602@researchut.com> Message-ID: On 18 February 2012 06:30, Ritesh Raj Sarraf wrote: > Recently I was viewing the fosdem video and the very same question was > asked there. It was told that the file system writes some metadata every > few seconds (IIRC, it was mentioned 30 seconds) but that that was tunable. Thanks, that explains a lot. (BTW, can I ask which video it was? I can only find one on snappr, not so much on btrfs) The 30 seconds agree well with what I'm seeing: spin-down after 20s, a couple of secs silence, then a spin-up, followed by a couple of writes, etc Would it be unfair to say this is a bug in btrfs, I mean the fact that it kind of ignores the laptop-mode state in the kernel? From rrs at researchut.com Sun Feb 19 17:08:20 2012 From: rrs at researchut.com (Ritesh Raj Sarraf) Date: Sun, 19 Feb 2012 21:38:20 +0530 Subject: [laptop-mode] laptop-mode with btrfs In-Reply-To: References: <4F3F4591.7060602@researchut.com> Message-ID: <4F411E74.3030704@researchut.com> This is an OpenPGP/MIME signed message (RFC 2440 and 3156) --------------enigE7D48C8907D7991FA4649B5D Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable On Sunday 19 February 2012 03:45 AM, Oei, Yung-Chin wrote: > Thanks, that explains a lot. (BTW, can I ask which video it was? I can > only find one on snappr, not so much on btrfs) http://www.youtube.com/watch?v=3DhxWuaozpe2I&feature=3Dg-hist&context=3DG= 26f0143AHTzz29AAVAA Somewhere in the video the audience ask the same question. --=20 Ritesh Raj Sarraf RESEARCHUT - http://www.researchut.com "Necessity is the mother of invention." --------------enigE7D48C8907D7991FA4649B5D Content-Description: OpenPGP digital signature Content-Disposition: attachment; filename="signature.asc" MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 900 bytes Desc: OpenPGP digital signature URL: --------------enigE7D48C8907D7991FA4649B5D-- From laptop-mode at yungchin.nl Mon Feb 20 00:19:07 2012 From: laptop-mode at yungchin.nl (yungchin) Date: Sun, 19 Feb 2012 23:19:07 +0000 Subject: [laptop-mode] laptop-mode with btrfs In-Reply-To: <4F411E74.3030704@researchut.com> References: <4F3F4591.7060602@researchut.com> <4F411E74.3030704@researchut.com> Message-ID: On 19 February 2012 16:08, Ritesh Raj Sarraf wrote: > On Sunday 19 February 2012 03:45 AM, Oei, Yung-Chin wrote: >> Thanks, that explains a lot. (BTW, can I ask which video it was? I can >> only find one on snappr, not so much on btrfs) > http://www.youtube.com/watch?v=hxWuaozpe2I&feature=g-hist&context=G26f0143AHTzz29AAVAA > > Somewhere in the video the audience ask the same question. Great, thanks, your memory is superb! If anyone else is looking for this: the discussion starts at ~29min; what's explained is that a few backups of the tree root are kept, and new backups are committed every 30s... regardless whether data changed or not. So now that I understand a little bit more, I suppose laptop-mode's changes to the various vm/dirty_* parameters can simply not affect this behaviour of btrfs, because it will write even with zero dirty pages.. From monnier at iro.umontreal.ca Tue Feb 28 03:21:39 2012 From: monnier at iro.umontreal.ca (Stefan Monnier) Date: Mon, 27 Feb 2012 21:21:39 -0500 Subject: [laptop-mode] Using laptop-mode on a desktop Message-ID: I'd like to tune my desktop so it consumes less power, especially when idle. I figure since my hardware (an AMD E350) is largely similar to the one of a laptop, I should be able to use the same approach, but while I'm perfectly willing to enable options that tradeoff a bit of performance for some energy saving, I'm not willing to trade off some much delayed write-back (with associated risk of data loss), nor aggressive spin-down (tho my Green drive can withstand a fair bit of it). Still, I should be able to reuse all the HDA, USB, SATA, Wifi, and other such power-management settings. Has someone tried that? I'd hope it even comes standard (basically split laptop-mode-tools into a part where the only tradeoff is performance, and a part that specifically deals with secondary storage, spin down, delayed write-back, mount options). I looked for such an option but couldn't find it (it seems that the HDD-twiddling is at the core and is the one thing that's harder to deactivate). Stefan From laptop-mode at yungchin.nl Tue Feb 28 17:00:19 2012 From: laptop-mode at yungchin.nl (yungchin) Date: Tue, 28 Feb 2012 16:00:19 +0000 Subject: [laptop-mode] Using laptop-mode on a desktop In-Reply-To: References: Message-ID: On 28 February 2012 02:21, Stefan Monnier wrote: > I looked for such an option but couldn't find it (it seems that the > HDD-twiddling is at the core and is the one thing that's harder to > deactivate). It's a bit workaround-ish, but would setting HD=" " in laptop-mode.conf (instead of the default line that's in there) do the trick? From rrs at researchut.com Tue Feb 28 20:03:28 2012 From: rrs at researchut.com (Ritesh Raj Sarraf) Date: Wed, 29 Feb 2012 00:33:28 +0530 Subject: [laptop-mode] Using laptop-mode on a desktop In-Reply-To: References: Message-ID: <4F4D2500.8000604@researchut.com> This is an OpenPGP/MIME signed message (RFC 2440 and 3156) --------------enig1F100727570BE4DD07F20CBF Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable This is exactly what laptop-mode-tools does. Almost everything is a plug-in here, including the Hard Driver settings. And everything is documented in the config files. CONTROL_HD_IDLE_TIMEOUT=3D0 CONTROL_HD_POWERMGMT=3D0 CONTROL_HD_WRITECACHE=3D0 CONTROL_MOUNT_OPTIONS=3D0 If you set these, your HDD should go untouched. By the way, what file system are you using? For ext3, the default commit intervals used to be every 5 seconds. In ext4, they introduced delayed allocation which would call an fsync() only when necessary. This led to big drama and people re-introduced a new option with the same ext3 behavior. So, please ensure your file system's behavior. Based on the distribution and the file system you use, the commit intervals might be different. On Tuesday 28 February 2012 07:51 AM, Stefan Monnier wrote: > I'd like to tune my desktop so it consumes less power, especially > when idle. I figure since my hardware (an AMD E350) is largely similar= > to the one of a laptop, I should be able to use the same approach, but > while I'm perfectly willing to enable options that tradeoff a bit of > performance for some energy saving, I'm not willing to trade off some > much delayed write-back (with associated risk of data loss), nor > aggressive spin-down (tho my Green drive can withstand a fair bit of > it). > Still, I should be able to reuse all the HDA, USB, SATA, Wifi, and othe= r > such power-management settings. Has someone tried that? I'd hope it > even comes standard (basically split laptop-mode-tools into a part > where the only tradeoff is performance, and a part that specifically > deals with secondary storage, spin down, delayed write-back, mount > options). > > I looked for such an option but couldn't find it (it seems that the > HDD-twiddling is at the core and is the one thing that's harder to > deactivate). --=20 Ritesh Raj Sarraf RESEARCHUT - http://www.researchut.com "Necessity is the mother of invention." --------------enig1F100727570BE4DD07F20CBF Content-Description: OpenPGP digital signature Content-Disposition: attachment; filename="signature.asc" MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 900 bytes Desc: OpenPGP digital signature URL: --------------enig1F100727570BE4DD07F20CBF-- From monnier at iro.umontreal.ca Wed Feb 29 14:25:38 2012 From: monnier at iro.umontreal.ca (Stefan Monnier) Date: Wed, 29 Feb 2012 08:25:38 -0500 Subject: [laptop-mode] Using laptop-mode on a desktop References: <4F4D2500.8000604@researchut.com> Message-ID: > This is exactly what laptop-mode-tools does. Almost everything is a > plug-in here, including the Hard Driver settings. And everything is > documented in the config files. > CONTROL_HD_IDLE_TIMEOUT=0 > CONTROL_HD_POWERMGMT=0 > CONTROL_HD_WRITECACHE=0 > CONTROL_MOUNT_OPTIONS=0 > If you set these, your HDD should go untouched. Thanks, that's helpful. But is that really all? What about the readahead, for example? Wouldn't it make sense to provide a CONTROL_HD that controls all of those options? > By the way, what file system are you using? ext3, mostly. Stefan From monnier at iro.umontreal.ca Wed Feb 29 14:55:58 2012 From: monnier at iro.umontreal.ca (Stefan Monnier) Date: Wed, 29 Feb 2012 08:55:58 -0500 Subject: [laptop-mode] Using laptop-mode on a desktop References: <4F4D2500.8000604@researchut.com> Message-ID: > Thanks, that's helpful. But is that really all? What about the > readahead, for example? Or the dirty-ratio settings? This said, I just tried your settings (together with ENABLE_LAPTOP_MODE_ON_AC=1), but there's one big problem left: ENABLE_AUTO_MODULES doesn't work because most/all modules set their LM_AC_* variables to basically deactivate the feature, so I'm back at enabling modules one by one and having to keep track of new modules when they appear. Why are those power-savings disabled when on AC? (especially given that ENABLE_LAPTOP_MODE_ON_AC itself is off by default) Other line of thought: the distinction I want to make here is between settings which affect exclusively performance and power. So for example, I don't want to use a 20s idle-timeout on my HD because desktop drives aren't designed for such aggressive spin-up/spin-down rates, and I don't want to delay write back further than the default, because in case of a crash it increases the amount of work lost. So based on this distinction, CONTROL_HD_POWERMGMT=1 should be OK because it only changes the performance of the drive but supposedly nothing else. Similarly, it should be OK to change the DIRTY_RATIO, DIRTY_BACKGROUND_RATIO and READAHEAD settings. But here comes another issue: IIUC these settings are mostly designed to let the drive spin down for longer periods of time, so while they're not directly problematic, these settings may not be of any use, powerwise, when the drive won't spin down anyway. So, they maybe should depend on CONTROL_HD_IDLE_TIMEOUT? Or do they already? What about MAX_LOST_WORK_SECONDS? Is it beneficial when CONTROL_HD_IDLE_TIMEOUT=0? Stefan From post at svennielsen.de Wed Feb 29 15:38:14 2012 From: post at svennielsen.de (Sven Nielsen) Date: Wed, 29 Feb 2012 15:38:14 +0100 Subject: [laptop-mode] Using laptop-mode on a desktop In-Reply-To: References: <4F4D2500.8000604@researchut.com> Message-ID: <4F4E3856.6090008@svennielsen.de> Fine-tuning laptop-mode-tools requires some moderate understanding of the underlying tools LMT uses for the performance/power tweaking. Read the manpage for hdparm, thoroughly from start to end. Read the description for tuning kernel parameters in /proc/sys/vm/ (which is what LMT also uses): http://www.kernel.org/doc/Documentation/sysctl/vm.txt That, and reading all the comments in LMT tools config files should give you a fairly good hang of how and why LMT does the things it does. Am 29.02.2012 14:55, schrieb Stefan Monnier: >> Thanks, that's helpful. But is that really all? What about the >> readahead, for example? readahead can be short or simply ignored if you have long spin down time. > Or the dirty-ratio settings? > > Why are those power-savings disabled when on AC? (especially given > that ENABLE_LAPTOP_MODE_ON_AC itself is off by default) Because you usually dont buy a fast computer to have it run slow. LMT is designed primarily for powersaving on laptops.. > Other line of thought: the distinction I want to make here is between > settings which affect exclusively performance and power. So for > example, I don't want to use a 20s idle-timeout on my HD because desktop > drives aren't designed for such aggressive spin-up/spin-down rates, and > I don't want to delay write back further than the default, because in > case of a crash it increases the amount of work lost. Then dont delay writeback and simply set a longer spin down time (e.g 600 for 10 minutes spin down), so you drives will either never spin down or when the PC is really doing nothing for a prolonged time. > So based on this distinction, CONTROL_HD_POWERMGMT=1 should be OK > because it only changes the performance of the drive but supposedly > nothing else. Set it to 0, as I assume you want max performance from your HD drives? > > Similarly, it should be OK to change the DIRTY_RATIO, > DIRTY_BACKGROUND_RATIO and READAHEAD settings. But here comes another > issue: IIUC these settings are mostly designed to let the drive spin > down for longer periods of time, so while they're not directly > problematic, these settings may not be of any use, powerwise, when the > drive won't spin down anyway. So, they maybe should depend on > CONTROL_HD_IDLE_TIMEOUT? Or do they already? > What about MAX_LOST_WORK_SECONDS? Is it beneficial when > CONTROL_HD_IDLE_TIMEOUT=0? See above, and, activate control HD idle timeouts if you want your drives to spin down (set longer intervals in config file as explained). Note that on a normal Linux Desktop setup many services constantly write out to disk, so you will have to do tweaking on the running services and different dirty settings to actually get your drives to spin down at all (use the 'powertop' utlity for analysing), IF you want spindown (as said, like when you leave the PC for, say, 1 hour). Again, tuning LMT requires some amount of expertise and knowing what you do. Regards, Sven From monnier at iro.umontreal.ca Wed Feb 29 17:28:57 2012 From: monnier at iro.umontreal.ca (Stefan Monnier) Date: Wed, 29 Feb 2012 11:28:57 -0500 Subject: [laptop-mode] Using laptop-mode on a desktop References: <4F4D2500.8000604@researchut.com> <4F4E3856.6090008@svennielsen.de> Message-ID: > Fine-tuning laptop-mode-tools requires some moderate understanding of > the underlying tools LMT uses for the performance/power tweaking. Yes, of course. But an important benefit of laptop-mode-tools is to make this largely unnecessary, by providing reasonable defaults. > Read the manpage for hdparm, thoroughly from start to end. I'm quite familiar with this manpage already. > Read the description for tuning kernel parameters in /proc/sys/vm/ > (which is what LMT also uses): > http://www.kernel.org/doc/Documentation/sysctl/vm.txt I haven't re-read it recently, but I don't think it's relevant to what I'm discussing. > That, and reading all the comments in LMT tools config files should give > you a fairly good hang of how and why LMT does the things it does. See below, we seem to be talking about different things. >>> Thanks, that's helpful. But is that really all? What about the >>> readahead, for example? > readahead can be short or simply ignored if you have long spin down time. We agree (I said "these settings are mostly designed to let the drive spin down for longer periods of time, so while they're not directly problematic, these settings may not be of any use, powerwise, when the drive won't spin down anyway"). Does laptop-mode-tools fiddle with readahead when CONTROL_HD_IDLE_TIMEOUT=0? If yes, why? If not, then it'd be great to make that connection more clear somewhere in the default laptop-mode.conf file. >> Or the dirty-ratio settings? >> Why are those power-savings disabled when on AC? (especially given >> that ENABLE_LAPTOP_MODE_ON_AC itself is off by default) > Because you usually dont buy a fast computer to have it run slow. Wrong assumption (in so many ways: for one, I bought a slow computer (even slower than the 4-year-old one it is replacing); for two, even if I had bought a fast computer I might care more about it being power-efficient when idle than about it being always running at top-speed; for three, even if (or especially because) it's slow I do care about its speed; for four, laptop-mode should not (and in my experience indeed does not) make a machine go slow). In any case, we're talking about the case where the user bothered to set ENABLE_LAPTOP_MODE_ON_AC=1, which is a pretty explicit statement from the user that even though he's running on AC, he wants to save power (because even though AC power is plentiful it is not without costs (e.g. impact on the machine's temperature and hence fan noise)). > LMT is designed primarily for powersaving on laptops.. Yes, and this thread is about using it in a different context. Think about it as "powersaving on nettops". Of course, we could start a new nettop-mode-tools, but the overlap with laptop-mode-tools is much too large to justify it: better improve laptop-mode-tools so it also covers the needs of desktop users who want to save energy. For those users who want big/easy switches, laptop-mode-tools currently offers: - the default, which is power-efficient on battery but not on AC. - ENABLE_LAPTOP_MODE_ON_AC=1 which has some odd behavior (set aggressive disk-spin-down, but don't bother activating the sata and HDA power management). and that's about it. And what I'm asking is - why is ENABLE_LAPTOP_MODE_ON_AC=1 the way it is, - a way for the user to say "I have desktop drives, so don't try too hard to spin them down". BTW, I guess the "don't try to spin down" probably also applies to flash disks. >> So based on this distinction, CONTROL_HD_POWERMGMT=1 should be OK >> because it only changes the performance of the drive but supposedly >> nothing else. > Set it to 0, as I assume you want max performance from your HD drives? Remember I have set ENABLE_LAPTOP_MODE_ON_AC=1, so no: I don't care that much about max performance from my drives. >> Similarly, it should be OK to change the DIRTY_RATIO, >> DIRTY_BACKGROUND_RATIO and READAHEAD settings. But here comes another >> issue: IIUC these settings are mostly designed to let the drive spin >> down for longer periods of time, so while they're not directly >> problematic, these settings may not be of any use, powerwise, when the >> drive won't spin down anyway. So, they maybe should depend on >> CONTROL_HD_IDLE_TIMEOUT? Or do they already? >> What about MAX_LOST_WORK_SECONDS? Is it beneficial when >> CONTROL_HD_IDLE_TIMEOUT=0? > See above, and, activate control HD idle timeouts if you want your > drives to spin down (set longer intervals in config file as explained). Spinning down desktop drives under a typical GNU/Linux desktop is pretty difficult in my experience: outside the reach of laptop-mode-tools. > Note that on a normal Linux Desktop setup many services constantly write > out to disk, so you will have to do tweaking on the running services and > different dirty settings to actually get your drives to spin down at all > (use the 'powertop' utlity for analysing), IF you want spindown (as > said, like when you leave the PC for, say, 1 hour). Exactly, laptop-mode-tools is not sufficient to get the drive to spin-down unless the spin-down delay is short (too short for a desktop drive). E.g. Firefox will write every half-hour his list of phishing sites. Stefan From rrs at researchut.com Wed Feb 29 21:05:37 2012 From: rrs at researchut.com (Ritesh Raj Sarraf) Date: Thu, 01 Mar 2012 01:35:37 +0530 Subject: [laptop-mode] Using laptop-mode on a desktop In-Reply-To: References: <4F4D2500.8000604@researchut.com> Message-ID: <4F4E8511.2060406@researchut.com> This is an OpenPGP/MIME signed message (RFC 2440 and 3156) --------------enig62E3FCE12A20B99F4A66C0C6 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable On Wednesday 29 February 2012 06:55 PM, Stefan Monnier wrote: >> CONTROL_HD_IDLE_TIMEOUT=3D0 >> > CONTROL_HD_POWERMGMT=3D0 >> > CONTROL_HD_WRITECACHE=3D0 >> > CONTROL_MOUNT_OPTIONS=3D0 >> > If you set these, your HDD should go untouched. > Thanks, that's helpful. But is that really all? What about the > readahead, for example? > > Wouldn't it make sense to provide a CONTROL_HD that controls all of > those options? > That'd definitely be nice to have. I might do it some time. If you want it now, feel free to provide a patch. --=20 Ritesh Raj Sarraf RESEARCHUT - http://www.researchut.com "Necessity is the mother of invention." --------------enig62E3FCE12A20B99F4A66C0C6 Content-Description: OpenPGP digital signature Content-Disposition: attachment; filename="signature.asc" MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 900 bytes Desc: OpenPGP digital signature URL: --------------enig62E3FCE12A20B99F4A66C0C6-- From rrs at researchut.com Wed Feb 29 21:17:43 2012 From: rrs at researchut.com (Ritesh Raj Sarraf) Date: Thu, 01 Mar 2012 01:47:43 +0530 Subject: [laptop-mode] Using laptop-mode on a desktop In-Reply-To: References: <4F4D2500.8000604@researchut.com> Message-ID: <4F4E87E7.7080702@researchut.com> This is an OpenPGP/MIME signed message (RFC 2440 and 3156) --------------enigEF635239CA12003424FD2047 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable On Wednesday 29 February 2012 07:25 PM, Stefan Monnier wrote: >> Thanks, that's helpful. But is that really all? What about the >> readahead, for example? > Or the dirty-ratio settings? > > This said, I just tried your settings (together with > ENABLE_LAPTOP_MODE_ON_AC=3D1), but there's one big problem left: > ENABLE_AUTO_MODULES doesn't work because most/all modules set their > LM_AC_* variables to basically deactivate the feature, so I'm back at > enabling modules one by one and having to keep track of new modules whe= n > they appear. > > Why are those power-savings disabled when on AC? (especially given > that ENABLE_LAPTOP_MODE_ON_AC itself is off by default) Enabling LMT on AC is not a common scenario. That setting is there because we do have users who use LMT on desktops and servers. For those, they should also tune every LM_AC_* setting. The default settings are keeping in mind the general user's use case. > Other line of thought: the distinction I want to make here is between > settings which affect exclusively performance and power. So for > example, I don't want to use a 20s idle-timeout on my HD because deskto= p > drives aren't designed for such aggressive spin-up/spin-down rates, and= > I don't want to delay write back further than the default, because in > case of a crash it increases the amount of work lost. > > So based on this distinction, CONTROL_HD_POWERMGMT=3D1 should be OK > because it only changes the performance of the drive but supposedly > nothing else. > > Similarly, it should be OK to change the DIRTY_RATIO, > DIRTY_BACKGROUND_RATIO and READAHEAD settings. But here comes another > issue: IIUC these settings are mostly designed to let the drive spin > down for longer periods of time, so while they're not directly > problematic, these settings may not be of any use, powerwise, when the > drive won't spin down anyway. So, they maybe should depend on > CONTROL_HD_IDLE_TIMEOUT? Or do they already? > What about MAX_LOST_WORK_SECONDS? Is it beneficial when > CONTROL_HD_IDLE_TIMEOUT=3D0? If TIMEOUT is 0, it would take whatever the defaults the driver sets. MAX LOST WORK SECONDS is for how long you would want your data in RAM provided you don't exceed the dirty ratio. --=20 Ritesh Raj Sarraf RESEARCHUT - http://www.researchut.com "Necessity is the mother of invention." --------------enigEF635239CA12003424FD2047 Content-Description: OpenPGP digital signature Content-Disposition: attachment; filename="signature.asc" MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 900 bytes Desc: OpenPGP digital signature URL: --------------enigEF635239CA12003424FD2047--