silencing a 2.5" drive under debian linux

[grandpa_boris]

my new fanless system is an EPIA-M in a morex 2699 case with a 2.5" laptop drive, seagate ST94011A. it's running debian linux and serves as my firewall/NAT/DHCP and whatever else i'll pile up on it in the future. being fanless, it's dead-silent -- except when the disk drive is on. the system has 512MB of RAM. i had set the disk to the most extreme power saving mode hdparm -B 1 /dev/hda , extreme noise management regime hdparm -M 128 /dev/hda , fairly aggressive shut down regime of 60 seconds hdparm -S 12 /dev/hda. but...

• the disk still gets accessed now and then, probably by various logging daemons. i searched these forums and the web for likely phrases that would catch info on disabling all disk-access activity by loggers, but found nothing relevant. does anyone remember if there was anything like this discussed here? does anyone have experience with software means of quieting linux boxes?

• the disk is listed as having reasonably benign noise characteristics, but in a silent machine with fairly unobstructed air flow this isn't silent. the drive has an annoying high pitch whine that would have been drowned by a fan, but is audible in a silent box. it's a much more annoying noise than what i get in another fanless EPIA-M system with seagate's 3.5" 7200.7 drive. the 2.5" drive is currently hard-mounted in a simple metal shell that comes standard in 2699. after ralf hutter's enthusiastic reports about using sorbothane, the i am considering remounting the disk using it, but will that kill the high-pitched noise?

one thing that had occured to me is that the drive is small enough, that i could create a sorbothane-choked enclosure for it. this is something i can't do with the 3.5" drive in this case because it seems to me there isn't enough space there to do it. with 2.5", may be i can somehow use the smalles zalman heatpipe to cool the drive?

comments?

[al bundy]

my new fanless system is an EPIA-M in a morex 2699 case with a 2.5" laptop drive... being fanless, it's dead-silent -- except when the disk drive is on... the drive has an annoying high pitch whine that would have been drowned by a fan, but is audible in a silent box... the 2.5" drive is currently hard-mounted in a simple metal shell that comes standard in 2699...

I've never used that Morex case, but is there any way you could place a SilentDrive enclosure in it? You might need to remove the standard metal cage, and find a creative place to mount the enclosure (perhaps using foam wedges). You can also remove the metal brackets from the SilentDrive enclosure, resulting in an even smaller enclosure to place within the computer case.

The SilentDrive enclosure will greatly silence the drive, and is compatible with any notebook harddrive. I'm saying this from my own personal experience - see my sig!

[grandpa_boris]

I've never used that Morex case, but is there any way you could place a SilentDrive enclosure in it?

i looked up the dimensions of the silentdrive enclosure, 6.5" x 5.5" x 1.5". i suspect that includes the side-brackets.

You can also remove the metal brackets from the SilentDrive enclosure, resulting in an even smaller enclosure to place within the computer case.

i didn't know those brackets were removable! quietpc has them on sale, it seems, and for under $30 i can take a shot at it. if it doesn't work, i expect i'll have other places i can use the enclosure. or it will make a nice present to someone .
websites that sell silentdrive don't mention laptop drives. how did you mount your drive in there? i haven't found any info about that googling the web, or the sites that sell it, or molex that makes it.

[Inexplicable]

i had set the disk to the most extreme power saving mode hdparm -B 1 /dev/hda , extreme noise management regime hdparm -M 128 /dev/hda , fairly aggressive shut down regime of 60 seconds hdparm -S 12 /dev/hda. but...

• the disk still gets accessed now and then, probably by various logging daemons. i searched these forums and the web for likely phrases that would catch info on disabling all disk-access activity by loggers, but found nothing relevant. does anyone remember if there was anything like this discussed here? does anyone have experience with software means of quieting linux boxes?

It's pretty hard quiet down the root disk. You probably have to mount it with the noatime option and disable all non-essential logging. Rather than do that, I toyed with the idea of putting the root file system and /var on a flash drive. Another idea might be to use ext3 in full data journaling mode and put the (large) journal on a flash drive. The latter might require modifying the way the journal gets written back on disk so that it only does it when the journal is getting full. Overall that might provide a very nice solution, as it would also work during active use. However, so far this is just idle musing as my HD is not really my biggest noise source.

[al bundy]

...i didn't know those brackets were removable!... how did you mount your drive in there?...

Thankfully the side brackets do come off rather easily, as they are attached with some small screws. I always remove them now, since mounting the SilentDrive with the brackets (in a 5.25" bay) causes some of the drive noise to transmit into the case - whereas using foam in a convenient case-interior location to jam-mount the enclosure is nice and quiet.

To mount the notebook drive in the SilentDrive enclosure, you'll want to first attach to the drive a standard notebook-to-IDE harddrive adapter, including those provided adapter brackets you see in the link picture. If you haven't used a SilentDrive enclosure before, give yourself a chance to try various drive and enclosure orientations to see which ones work best for you. You'll find the enclosure to be surprisingly easy to work with and allowing for many different orientations and mounting alternatives. So long as you stick with notebook drives, or with standard IDE drives of 5400rpm or less, you won't have any serious heat issues either.

Good luck! I hope it works well for you.



Edit: Added comment below...

[JonW]

Someone had a thread about "quiet filesystems" a while back.

I think the conclusion was that you'll have problems getting your disk to shut down using a journalling file system and hdparm. Have a google for noflushd, a daemon which prevents flushing to disk, and allows the drive to shut down. Apparently it works best (only?!) with a non-journalled FS, such as ext2.

I'm thinking of putting Gentoo on my PVR, and using this to keep the disk quiet the majority of the time. Anyone know if ext2 can handle large parititions well (i.e. about 120GB?)

[grandpa_boris]

Have a google for noflushd, a daemon which prevents flushing to disk, and allows the drive to shut down. Apparently it works best (only?!) with a non-journalled FS, such as ext2.

i have noflushd installed. i don't think it has had substantive effect, since i do have ext3 on /. the disk turns off, stays off for a while, then comes back on as something pushes it to write something on the disk. i am worried that the large number of starts and stops may shorten the disk's life span as effectively as the excessive heat could...

I'm thinking of putting Gentoo on my PVR, and using this to keep the disk quiet the majority of the time. Anyone know if ext2 can handle large parititions well (i.e. about 120GB?)

i think the suggestion from al bundy, above, about placing the disk in a silentdrive enclosure is a simpler solution than anything else i had contemplated.

[tay]

Do you have noatime set in your fstab?

Code: Select all

/dev/hda1       /               ext3    defaults,noatime 

You can also use tune2fs -c to change the commit interval but that is risky IMO. You could just go back to ext2 since a high commit interval fscks the journalling capability anyway.

[grandpa_boris]

Do you have noatime set in your fstab?

Code: Select all

/dev/hda1       /               ext3    defaults,noatime 

i will now . that should help with access to cached inodes. i expect that realistically, the system should almost never look at the disk. even with some memory stolen by the video chipset, it leaves 483MB for the system use, of which 273MB are free right now, after 2 days of being up. i expect that most of everything it needs is now in the caches. certainly doing "find / -print | wc -l" caused it to access the disk only once at the end, probably to update atime.

You can also use tune2fs -c to change the commit interval but that is risky IMO. You could just go back to ext2 since a high commit interval fscks the journalling capability anyway.

the man page of tune2fs on my box (which is an almost latest debian) doesn't seem to have any options to reduce the commit intervals. it will let me disable journaling, but the filesystem should almost never be written and i come from the environment where intent logging is not just the religion, but a way of life , so i hesitate to disable logging. hopefully between noflushd and your suggestion of "noatime" mount option, the fs will never need to access the disk.
i will still experiment with silentdrive, just to find out if it will fit. that way if i ever decide to do something more intensive with the remaining 34GB of space on this 40GB disk, the noise will not be a problem.

[tay]

Youre right, I was mistaken as to what was possible with tune2fs.

[al bundy]

Hey grandpa_boris, just another comment here...

When fitting the SilentDrive in the case, while it can be helpful in some situations to remove the side brackets in a SilentDrive enclosure (the ones that are attached with screws), I wouldn't recommend also removing the main brackets (i.e. the 'metal-plate' ones that slide in/out of the enclosure unit).

The reason I say this is that while the side (screw-on) brackets don't contribute to heat transfer in the SilentDrive enclosure, the main (slide in/out) brackets are designed to transfer heat from the drive to the external enclosure environment.

So don't go too far! I realize that you probably already figured this, but I thought I'd add it anyway just to be careful...

[grandpa_boris]

Hey grandpa_boris, just another comment here...

When fitting the SilentDrive in the case, while it can be helpful in some situations to remove the side brackets in a SilentDrive enclosure (the ones that are attached with screws), I wouldn't recommend also removing the main brackets (i.e. the 'metal-plate' ones that slide in/out of the enclosure unit).

i still haven't gotten the silentdrive enclosures, so i hadn't had a chance yet to make that mistake.

one thing that puzzles me: with the 3.5" drives, it seems the drive is simply inserted in and the fit is very tight. the thermal contact happens because it can't be avoided. there are no screwes used to hold the drive in. so i am still wondering how to anchor the 2.5" drive in there. sorbothane? that will not cause the contact to be thermally "useful". 3M high thermal transfer sticky tape? i haven't yet even tried finding it in stores. any suggestions on where to find that?

[al bundy]

...there are no screwes used to hold the drive in. so i am still wondering how to anchor the 2.5" drive in there...

When using the adapter and its brackets that I mentioned in the above post, you get a tight enough fit in the enclosure without any need whatsoever for further drive anchoring. The IDE and power cables also make good contact with the foam insert of the enclosure back-cover, to prevent any sliding of the drive in the enclosure. It's not going anywhere after it is all put together properly.

The metal plates in the enclosure easily absorb and transfer the small amount of heat that a notebook drive releases - no need for direct thermal contact.

Works like a charm!

[Tigr]

Check out the use of tmpfs. You could make a disk if you have enough memory to put all the files normally accessed thorugh /var and /tmp onto. Makes for way less accesses.

[grandpa_boris]

Check out the use of tmpfs. You could make a disk if you have enough memory to put all the files normally accessed thorugh /var and /tmp onto. Makes for way less accesses.

i sort of understand the use of tmpfs, but not how to use it with /var and i am not sufficiently familiar with the logging aspects of linux to know what needs to be tmpfs'd. which parts of /var do you suggest be put up as tmpfs? the ones i see people use tmpfs for are /var/lock and /var/run. what else?

i also see some people put the entire /var as tmpfs and just mkdir /var/run, /var/lock and /var/log, but it looks like debian packages install a whole lot of stuff into /var, over 256M (most of it is the package cache, it seems).

i will experiment with at least placing /var/lock and /var/run. i expect that overmounts work without any issues on ext3.... i am not brave enough to put /var/log into tmpfs and i don't think i want to try to play games with copying logs somewhere, umounting /var/log, then copying the logs back to the now-visible real /var/log.

[grandpa_boris]

while googling for "debian tmpfs mount" i found a description of a much better targetted appliance build than what i have: no disks, just 32MB of flash. given that a 256MB CF card costs $50, i think it's reasonable to be a bit more profligate with the configuration than this guy had to be a year ago. and we have here a local, reasonably priced source of compact and reasonably priced fanless, low-powered appliance platforms... hmmm. this may become my next silent project...

[Tigr]

Actually, I do not think that placing /var/log on tmpfs is dangerous. In the worst case, you loose your logfiles. So what? You can write a script that will place the log files into some location on your hard drive instead of dumbly renaming them to logfile.1 and replace 'savelog' with that script in cron tasks. In the case of a crash, you lose the last day of logfiles. I run the check scripts on my logfiles every 5 minutes to check for break-in attempts and general errors. Anything suspicious is mailed to me right away. If you have that, basically no important information will be missing unless something happens really quickly.

You are right, there is a whole lot of stuff in /var but you do not need to place all of it on tmpfs. All you need is probably /var/run, /var/lock, /var/tmp, /var/log and /tmp. Those are probably the most often written. Actually, the processes do not stop/start that often on a firewall machine so I think just moving /var/log will allow you machine's drive to sleep most of the time. And the firewall usually stays up for a long time without problems due to the absence of lusers so I do not see a problem with using tmpfs in this way.

EDIT:
As to how to use it actually. I normally mount /tmp on tmpfs
/tmp /tmp tmpfs defaults,auto 0 0
and then just soft link the files I need (like ln -s /tmp/log /var/log).

While looking at the link you provided I noticed that they use tmpfs in the same fashion. Check the section "Mount in bind mode".

[grandpa_boris]

i finally got the silentdrive enclosure and got around to trying to fit it in. unfortunately, the case simply doesn't accomodate it! in morex 2699, even when stripped off of all external metal parts, it needs perhaps 1/4" more space to fit. it can't get into the case because the motherboard gets in the way.

time to buy a sheet of sorbothane, build a box like silentdrive, but smaller, line it with sorbothane, cut a hole for a mounting metal plate that will form the cooling "fin", mount the disk drive to this fin and hope that i get all holes to line up...

if that ends up being too daunting a challenge, i'll look into replacing the drive with a flash card and creating a much smaller configuration that will easily fit into the 64MB card i have had sitting on my desk for the last year...

lesson learned: just because the dB numbers look small doesn't mean that the noise will not be very irritating, despite not being loud.

[grandpa_boris]

As to how to use it actually. I normally mount /tmp on tmpfs
/tmp /tmp tmpfs defaults,auto 0 0
and then just soft link the files I need (like ln -s /tmp/log /var/log).

mount shows

Code: Select all

/dev/hda6 on / type ext3 (rw,noatime,errors=remount-ro)
proc on /proc type proc (rw)
devpts on /dev/pts type devpts (rw,gid=5,mode=620)
/dev/hda1 on /boot type ext3 (rw,noatime)
tmpfs on /tmp type tmpfs (rw,noatime)
tmpfs on /var/run type tmpfs (rw,noatime)
tmpfs on /var/log type tmpfs (rw,noatime)
tmpfs on /var/lock type tmpfs (rw,noatime)
tmpfs on /var/tmp type tmpfs (rw,noatime)

i also set the disk to the fastest possible spin down with hdparm -S 1 /dev/hda and the most aggressive power management. never the less, every so often the system still spins up the disk for no obvious reason. could it be something awfully persistent in ext3? flash disks are beginning to look more and more appealing...

[Unseaming Chris]

Have you had any luck physically making the hard drive quieter? I was planning on using the ST94011A in very similar sytem as a mp3 player but the noise problem worries me. Is the 3.5" 7200.7 really better?

[grandpa_boris]

Have you had any luck physically making the hard drive quieter?

not yet. because silentdrive enclosure will not fit in the case, i will next be trying to use sorbothane to line inside and out the metal enclosure holding the drive to dampen the high frequency squeal. this machine is currently running as my gateway/router/firewall, so taking it out of service is always a hassle, making the progress on this project very slow.

I was planning on using the ST94011A in very similar sytem as a mp3 player but the noise problem worries me. Is the 3.5" 7200.7 really better?

i have two identical morex mini-ITX boxes, one with 7200.7 (running w2k) sitting 2 feet in front of me, the other with the 2.5" ST94011A (running debian), is 6' away. i can't hear 7200.7 even when it's running a virus scan (unless i put my ear right on the case or right next to the vent grilles). i can plainly hear ST94011A's whine even when it's not doing anything, just spinning in the active/idle mode. the arm movement noise on ST94011A is also quite audible and "clickety". on the other hand, 7200.7 runs at 54'C in this enclosure when loaded, 51'C when idle (with the annoying 40mm fans on at 7-8V, it runs at 43'C idle, 48'C loaded). ST94011A runs at 31'C when loaded, 26'C when idle. since the systems are just 6' apart and at the same height, i have to believe that the ambient temperature is almost the same.

so there's my choices: silence and heat death vs. a low power and low temperature operation with annoying squealy noise and convoluted attempts to silence the drive.

if you are planning to use a larger enclosure without cooling problems or can use an 80mm fan at low RPMs (which will be inaudible if you use panaflo L1A or similar high quality fan), i'd most definitely get the 7200.7 or the newer samsung spinpoints that have been getting very positive reviews in these forums.

[Unseaming Chris]

Thank you very much for the information, probably saved me years of annoyance. The 3.5" drive sounds too hot for 24-7 usage so I think I'll look into slower 2.5" HDs like the Fujitsu MHS2040AT. Speed doesn't really matter for just playing mp3s.

[grandpa_boris]

this thread shows a silent enclosure i haven't seen before. another article in the same thread references the listing of the same enclosure on a german site that also shows its dimensions. it's 100x140x30mm.

silentdrive, according to directron.com, is 6.5" x 5.5" x 1.5" (165x140x38mm). without the cooling fins and if the screw mounts were filed off, the silentdrive enclosure is still around 115-116m wide. morex 2699 leaves a little less than 4.5" or 114mm of space for the disk drive. this is plenty to mount a 3.5" drive (~89mm), but not enough to fit in the silentdrive box. but the silentmaxx enclosure looks like it should fit (as long as the first number is really width, rather than length!:shock:)

[dukla2000]

Just remembered this thread while I have been playing with Knoppix the past 24 hours.

Basically, somehow, Knoppix manages to eliminate disk seeks. As it is CD-ROM only disk access are distinctly noticable But I have set a new box folding (no hdd, only a spare CD and pendrive) and the CD hasn't spun up for hours now. Only if I move the mouse or whatever.

Can barely spell Linux myself so wouldn't know where to start looking to find what Knoppix has configured to get to this state, but whatever it is, it works.

[Inexplicable]

It's simple enough. The CD is read-only, so it never spins up because of disk writes. All those things that would normally be written to disk are either disabled or directed to a ram filesystem. The file system cache takes care of optimizing the reads. The more system ram you have the better. In principle you can easily do the same thing with a hard disk, if you don't care about logging system events or saving your work.

[mgiammarco]

There is a kernel patch called laptop mode that:
- tries to keep hard disk powered down.
- optionally log software that ask frequently to spin up so you can detect them and remove/change

[grandpa_boris]

There is a kernel patch called laptop mode that:
- tries to keep hard disk powered down.
- optionally log software that ask frequently to spin up so you can detect them and remove/change

it appears the laptop_mode is already integrated into current versions of linux kernels. i picked up the laptop_mode.sh script and changed it to spin down after 5 seconds of idle time instead of 25 seconds (hdparm -S1 instead of -S5 or 6 it was). the increased buffer flush period helps somewhat. the drive stays spun down a lot longer now. which is still not quite right, because ideally that system shouldn't be flushing at all! it's a gateway passing IP traffic. the only time it should be writing anything is when the DHCP leases expire and are renewed by the clients, which happens every few hours. at that point it's supposed to log the new lease info. the rest of the time it has no dirty pages to flush and no swapping to do (it's 512MB ram for a system that can easily run in 64-128MB).

the noise of the drive, when it starts up, is still annoying, though. MikeC's recent review of e-Otonashi points out that the seagate momentus drive (which is what i have) has an annoying whine despite his well-known skill in silencing disks. the fujitsu drives he also tried don't have this issue. it's unfortunate that the review was published after i built this machine. i will consider replacing the drive with fujitsu one, or try building a miniature version of silentdrive out of excessive amounts of sorbothane.

[grandpa_boris]

Just remembered this thread while I have been playing with Knoppix the past 24 hours.

thanx, dukla2000. i took a look at knoppix and it looks interesting. the problem i have with running it is that this system is my gateway/firewall and i am trying to keep it as current and as up to date on security patches as possible. this is a trivial matter with the system i have now, because it can update itself restart key components without shutting down the primary functionality. with knoppix, this would require frequent re-burning if the CDROM and reboots of the system. it's doable, but a lot more process-intensive.

[Pascal]

Hi grandpa_boris,

I had a problem similar to yours a few months ago and the following page helped me a lot

http://bulma.net/body.phtml?nIdNoticia=1511

I hope it will be useful to you too.

[grandpa_boris]

Hi grandpa_boris,

I had a problem similar to yours a few months ago and the following page helped me a lot

http://bulma.net/body.phtml?nIdNoticia=1511

I hope it will be useful to you too.

interesting stuff, pascal, and i would have never found this myself!
i'll try this guy's suggestions, but my expectations are low: all of the logs mentioned in the article are already on tmpfs mounts (so they aren't written to disk), all mounts are noatime, and because this box runs as a pass-through gateway it isn't running news or mail daemons. it's possible that writing to /dev/null rather than tmpfs will make the key difference....