[SOLVED] large file copy hangs using bfq on jfs (linux-ck and -lqx)

sitquietly · 2011-10-25 22:05:50

In the past couple of days I've found on my system that with linux-ck-3.0.7 or linux-lqx-3.0.6, using the bfq block i/o scheduler, a large file copy operation will hang and never complete if the target partition is a jfs file system.

With either the ck or liquorix kernel, if the bfq scheduler is used, I can't copy a 6.8 GB iso from an ext4 partition, or from another machine on the network, to a jfs partition. It copies as much as 5 GB before hanging. After the copy operation hangs if root performs a "sync" or "reboot", the sync or reboot operation never terminates and the system can be recovered only by a forced reboot (reboot -f) from another terminal. There weren't any error messages in the system logs. The operation just never completes.

The same iso can be copied successfully TO the ext4 partition.

This is happening on an i5 Sandy Bridge system with 16 GiB of ram. The target drive is a rotational hard drive, a Samsung HD103si. The target partition is 888 GB. The kernels were compiled for a core2 or corei7-avx target with bfs task scheduler at 1000 Hz. Neither drive nor jfs file system exhibits any other problems. The file copies perfectly if I use the cfq elevator. I think that maybe there is a bad interaction between the bfq scheduler and jfs file system in recent kernels. I don't know which release first showed this problem; I've had mysterious problems for maybe a week, such as inability to get the computer to shutdown normally. I think it is only in the 3.0.x kernels.

For now I'm going back to the cfq i/o scheduler.

Can anyone confirm this problem with the bfq i/o scheduler on other systems?

[EDIT] I've confirmed this problem on another machine. The 6.8 GB iso can not be copied onto a jfs partition on an IONITX Atom 330 system (i.e. NOT a Sandy Bridge system). Again, it copies properly using the cfq scheduler. It seems to me that the bfq patch can not be used with the jfs file system with recent kernels.

Please make a reply here if you CAN copy large files onto jfs with any system using the bfq scheduler on 3.0.x kernels.

Last edited by sitquietly (2012-04-09 21:17:07)

sitquietly · 2012-04-09 21:15:54

sitquietly wrote:

In the past couple of days I've found on my system that with linux-ck-3.0.7 or linux-lqx-3.0.6, using the bfq block i/o scheduler, a large file copy operation will hang and never complete if the target partition is a jfs file system.....Can anyone confirm this problem with the bfq i/o scheduler.....

This problem has been resolved. Arianna Avanzini wrote to me on behalf of the BFQ developers and reported that "the BFQ scheduler had an important bugfix release (BFQ-v3r1) which corrected a request dispatch bug that was introduced when releasing BFQ-v3 for the 3.0.x kernel version. This bug may have caused the issue You observed, as it was found to produce occasional I/O hangs even with ordinary file systems." That bug has been fixed.

I've tested the large file copies that previously could "reliably" hang my system. Using the latest linux-ck (3.2.1-ck) kernel, with the BFQ scheduler, there is no problem. Much thanks to the BFQ devs! BFQ gives a noticeable improvement on my systems and its good to know that it's reliable again

Arianna wants us to know that "BFQ has its own mailing list for user issues and developer announces (https://groups.google.com/forum/?fromgr … fq-iosched): if You have any problems to report, please feel free to send a mail there. The author (Paolo Valente) and the other developers will be glad to help."

graysky · 2012-04-09 21:18:28

Latest incarnation of Linux-ck ships with BFQ-v3r3 for 3.3. You need to freshen your kernel more often

sitquietly · 2012-04-09 22:50:34

graysky wrote:

Latest incarnation of Linux-ck ships with BFQ-v3r3 for 3.3. You need to freshen your kernel more often

Yes, thank you, and thanks very much for your work on linux-ck -- I like it a lot and keep both linux-ck and linux-lqx up-to-date on my computer. I've been keeping up-to-date with your releases but just yesterday got confirmation from the BFQ devs that the problem I had "long long ago" with 3.0.x kernels was confirmed to be a bug in BFQ and was indeed fixed. The report of the fix is for the sake of folks coming to this thread from google searches. There's nothing to see here folks. It's been fixed. Move along.

Arch Linux

#1 2011-10-25 22:05:50

[SOLVED] large file copy hangs using bfq on jfs (linux-ck and -lqx)

#2 2012-04-09 21:15:54

Re: [SOLVED] large file copy hangs using bfq on jfs (linux-ck and -lqx)

#3 2012-04-09 21:18:28

Re: [SOLVED] large file copy hangs using bfq on jfs (linux-ck and -lqx)

#4 2012-04-09 22:50:34

Re: [SOLVED] large file copy hangs using bfq on jfs (linux-ck and -lqx)

Board footer