Weekly edition	Kernel	Security	Distributions	Contact Us	Search
Archives	Calendar	Subscribe	Write for LWN	LWN.net FAQ	Sponsors

Linux Filesystem, Storage, and Memory Management Summit, Day 1

Posted Apr 15, 2011 21:39 UTC (Fri) by oak (guest, #2786)
Parent article: Linux Filesystem, Storage, and Memory Management Summit, Day 1

"Part of the problem is probably that the filesystem (ext3, in particular) will end up blocking processes which are waiting on buffers until a big journal commit frees some buffers. That can cause writes to a slow device to stall unrelated processes."

Unrelated = processes trying to read unrelated data from FS?

"It seems that there is more going on, though,"

Things like apps doing sync, or huge writes filling page cache and resulting low memory condition causing other processes data to be swapped out (and immediately back in, if they are active processes)?

(Log in to post comments)

HUGE regression! Vital!

Posted Apr 21, 2011 10:26 UTC (Thu) by blujay (guest, #39961) [Link]

IMHO this problem should be the number one priority for all fs devs until it's completely fixed.

I've used Debian/Ubuntu full-time since 2003, and using FAT32 USB flash drives was never a problem until a few years ago.

As it stands now, writing files more than a few hundred megs to a FAT32 USB drive in Linux is barely possible, unless you have hours and hours to wait! Windows has no problems with this at all, and older kernels didn't either. This seems like the biggest regression, ever, to me. FAT32 is necessary for writing to USB drives and camera memory cards and exchanging data between different OSes--it's crucial!

What boggles my mind is that it's such a low priority for the kernel devs--that it's been a bug for so long. Don't they use USB drives? Don't they exchange data with Windows users, ever? Do they really use ext3/4 on all their USB flash drives? (Remember that most USB flash drives are optimized in hardware to work better with FAT32 by handling the FAT blocks differently.)

How can we expect people to take Linux seriously if we have to tell them that it can't write 200MB to a USB drive in less than an HOUR?! I think Windows 98 could do better than that!

HUGE regression! Vital!

Posted May 9, 2011 13:22 UTC (Mon) by nix (subscriber, #2304) [Link]

I regularly write >500Mb to FAT-formatted USB drives and I certainly don't see this behaviour. Is this USB 1 or something?

HUGE regression! Vital!

Posted Oct 26, 2011 21:20 UTC (Wed) by dougg (subscriber, #1894) [Link]

I see "gone to lunch" type performance from commodity SD cards that contain a root file system when packages are installed. The mmcqd/0 kernel process seems to hang on IO, clearing up to several minutes later! That is with SD cards from the biggest names in that field. However if I use an industrial grade SD card (e.g. from ATP and a lot more expensive) then I don't see that lousy performance. USB flash cards would probably suffer from the same problem; and USB adds broken error processing on top of horrible mass storage class implementations. Are there industrial grade USB flash cards?