usb disks?? Automounting on /media once, slow writing

David Zeuthen david at fubar.dk
Sun May 22 21:37:00 UTC 2005


On Sun, 2005-05-22 at 15:25 -0600, Michal Jaegermann wrote:
> On Sun, May 22, 2005 at 10:08:21PM +0200, René van Paassen wrote:
> > Inserted a usb 2.0 disk, and the thing duly shows up on /media/usbdisk
> > However, writing was painfully slow, order of 100 kB/s
> 
> That is most likely because hal is mounting removable USB devices
> with 'sync' option.  Personally I think that this is insane and
> will cause numerous complaints and claims that Linux is extremely
> slow but this is supposed to be "idiot-proof".
>
> Mounting sync is really bad for a writing speed, and likely quite
> bad for "USB sticks" even if they do "wear levelling" in a
> reasonable way, but gives you a fighting chance that you did not
> screwed up your data and/or a file system on that device if you
> pulled it out without unmounting first.  Mind you - only a chance
> and very far from guarantees.  Of course if you will kill your USB
> memory stick in the process this is not much of a gain.
> 
> There is some possibility that you can change that policy on your
> machine if you will be able to figure out how - which is hard to
> call "documented".
> 

The next build will not use the 'sync' option, mainly because of
performance reasons (now that O_SYNC works for vfat). See also

 https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=157674

> > Tried a second disk, this one would not automount. After manually
> > mounting it (/dev/sde1) it ran well, with normal writing speed. 
> 
> A default for mount is 'async' so you will see what you see unless
> you asked otherwise.
> 
> > Inserted the original disk again, and would not automount. After manual
> > mount (/dev/sde this time) it also ran well. 

This is due to a kernel bug which mixes up the order of hotplug events
and does other things that confuse userland in interesting ways.

 https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=156167

David





More information about the test mailing list