[Lustre-discuss] Lots of Lustre Errors with Lustre 1.6.4.2

[Johann Lombardi]

On Mon, Feb 18, 2008 at 08:57:32AM +0100, Dilling wrote:
> where xxx is one of the following: commitrw,journal start,parent  
> lock,preprw_write,start_page_write,journal start,parent lock,preprw_read 
> setup,start_page_read,setattr
> and xy is of order 10-40s. The frequency of these messages depends on  
> the load of the OSTs. On all OSTs we observe a very high load between 20 
> and 120 alternating between all 4 OSTs. We also see up too 900 ll_ost_io 
> threads.

Have you seen 900 ll_ost_io threads on a single OSS?
Theoretically, the number of ll_ost_io is limited to 512 (i.e. OSS_THREADS_MAX).
Could you please try to enforce the number of ost threads to 128 or 256?
To do so: stop the OSTs, unload the lustre modules and add the following line in
/etc/modprobe.conf:
options ost oss_num_threads=128

Johann