On 12/12/11, Stephen Hemminger <[email protected]> wrote:
> On Mon, 12 Dec 2011 06:43:45 +0800
> Barry Kauler <[email protected]> wrote:
>
>> On 12/7/11, Thomas Himmelmann <[email protected]> wrote:
>>
>> > Unionfs 2.5.10 don't work stable on kernel 3.0 and 3.1 for me.
>>
>> 2.5.10 is rock-solid with a 3.0.7 kernel for me.
>>
>> There is something that I found out about the Linux kernel, that might
>> affect the stability of Unionfs...
>>
>> A little while ago, we found that one of these config option makes
>> operation unreliable:
>>
>> CONFIG_SCHED_SMT
>> SMT scheduler support improves the CPU scheduler's decision making
>> when dealing with Intel Pentium 4 chips with HyperThreading at a cost
>> of slightly increased overhead in some places. If unsure say N here.
>>
>> CONFIG_SCHED_MC
>> Multi-core scheduler support improves the CPU scheduler's decision
>> making when dealing with multi-core CPU chips at a cost of slightly
>> increased overhead in some places. If unsure say N here.
>>
>> We were getting mysterious errors when using the 'cp' command. When
>> copying a complete directory hierarchy, some files/directories were
>> sometimes completely missing from the destination. There were no error
>> messages. This has nothing to do with Unionfs (these errors were
>> occurring in ext2/3 partitions outside of any layer filesystem), I am
>> describing a fundamental flakyness of the kernel, not just the 3.x,
>> earlier too.
>>
>> I discovered that turning off CONFIG_SCHED_SMT and CONFIG_SCHED_MC
>> fixed it. It is probably only one of those that is the culprit though,
>> probably CONFIG_SCHED_MC, but I never verified that.
>>
>> So, my kernel has:
>>
>> CONFIG_SMP=y
>> # CONFIG_SCHED_SMT is not set
>> # CONFIG_SCHED_MC is not set
>>
>> I recommend, configure your kernel the same, then see if you still get
>> the problems with Unionfs 2.5.10.
>
> If those fixed your problems, then it means locking
> is borked. I.e not a scheduler problem. What kind of device driver?
> Could be bad disk controller driver. Did you report it to distro or kernel
> mailing list?
>

It is getting off-topic, but no, the problem was not with one
computer. Different kernel; versions, different computers, even a
brand new laptop.

As it happened seemingly randomly, I could not reproduce it, and did
not report it to lkml. Other Puppy Linux testers reported the same
problem.

Actually, ever since I first started using an SMP-enabled kernel with
Puppy Linux, I left those two parameters disabled. It was only early
this year that I enabled them, and we started hitting this random
error problem. I disabled them again, and all is well.

Anyway, this is o.t. I only brought it up, curious if it could be a
factor in errors some people are reporting with latest Unionfs. If
said person thinks that is highly unlikely, then they are welcome to
ignore my suggestion.

Regards,
Barry Kauler
_______________________________________________
unionfs mailing list: http://unionfs.filesystems.org/
[email protected]
http://www.fsl.cs.sunysb.edu/mailman/listinfo/unionfs

Reply via email to