[Lustre-discuss] [zfs-discuss] ZFS/Lustre echo 0 >> max_cached_mb chewing 100% cpu
Andrew Holway
andrew.holway at gmail.com
Tue Oct 22 09:43:31 PDT 2013
On 22 October 2013 16:21, Prakash Surya <surya1 at llnl.gov> wrote:
> This probably belongs on the Lustre mailing list.
I cross posted :)
> Regardless, I don't
> think you want to do that (do you?). It'll prevent any client side
> caching, and more importantly, I don't think it's a case that's been
> tested/optimized. What're you trying to acheive?
Sorry I was not clear, I didn't action this and I cant kill the
process. It seemed to start directly after running:
"FSTYPE=zfs /usr/lib64/lustre/tests/llmount.sh"
I have tried to kill it first with -2 upto -9 but the process will not budge.
Here is the top lines from perf top
37.39% [osc] [k] osc_set_info_async
27.14% [lov] [k] lov_set_info_async
4.13% [kernel] [k] kfree
3.57% [ptlrpc] [k] ptlrpc_set_destroy
3.14% [kernel] [k] mutex_unlock
3.10% [lustre] [k] ll_wr_max_cached_mb
3.00% [kernel] [k] mutex_lock
2.82% [ptlrpc] [k] ptlrpc_prep_set
2.52% [kernel] [k] __kmalloc
Thanks,
Andrew
>
> Also, just curious, where's the CPU time being spent? What process and/or
> kernel thread? What are the top entries listed when you run "perf top"?
>
> --
> Cheers, Prakash
>
> On Tue, Oct 22, 2013 at 12:53:44PM +0100, Andrew Holway wrote:
>> Hello,
>>
>> I have just setup a "toy" lustre setup using this guide here:
>> http://zfsonlinux.org/lustre and have this process chewing 100% cpu.
>>
>> sh -c echo 0 >> /proc/fs/lustre/llite/lustre-ffff88006b0c7c00/max_cached_mb
>>
>> Until I get something more beasty I am using my desktop machine with
>> KVM. Using standard Centos 6.4 with latest kernel. (2.6.32-358.23.2).
>> my machine has 2GB ram
>>
>> Any ideas?
>>
>> Thanks,
>>
>> Andrew
>>
>> To unsubscribe from this group and stop receiving emails from it, send an email to zfs-discuss+unsubscribe at zfsonlinux.org.
>
> To unsubscribe from this group and stop receiving emails from it, send an email to zfs-discuss+unsubscribe at zfsonlinux.org.
More information about the lustre-discuss
mailing list