[lustre-discuss] Real cases where setstripe is useful

Laifer, Roland (SCC) roland.laifer at kit.edu
Wed Mar 16 02:47:00 PDT 2016


Andreas,

I recently learned that setting a default layout on the filesystem root 
right after the filesystem was created has on big disadvantage: This 
information is inherited to subdirectories and if you want to change the 
default later this does not affect existing subdirectories. Therefore, 
it might be better to set useful parameters at file system creation time 
or use the complicated way to change the file system configuration.

We had one use case for this. Several users reported I/O errors because 
of the short read problem (LU-6389) which was discussed here a while 
ago. The workaround was to set the default stripe count to 1. This was 
easy for file systems where we had set the stripe count at filesystem 
creation time and caused trouble for file systems where we had changed 
the stripe count on the filesystem root.

Regards,
   Roland


Am 15.03.2016 um 23:41 schrieb Dilger, Andreas:
> From: Jeevan Behara Patnaik (GIS)
> Sent: 15 March 2016 12:49
> To: lustre-discuss at lists.lustre.org<mailto:lustre-discuss at lists.lustre.org>
> Subject: Real cases where setstripe is useful
>
> Hi,
>
> I have been reading about using setstripe exclusively for files or directories. I would like to know if it is helpful for a grid jobs.
>
> We are going to use lustre in grid environment in place of NFS. So, at what stage, can I exactly use setstripe?
>
> These jobs would generate files and the creation of those files are not in our control and the size to my knowledge, may not be determined in prior.
>
> So, does it mean that in grid, we could only use the default stripe for all files and directories?
>
> The best way to handle this is to set a default layout on the filesystem root that suits most applications, and then set a striping on the output directories for applications that have different file IO requirements.
>
> Cheers, Andreas
> --
> Andreas Dilger
> Lustre Principal Architect
> Intel High Performance Data Division
>
>
>
> _______________________________________________
> lustre-discuss mailing list
> lustre-discuss at lists.lustre.org
> http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org



More information about the lustre-discuss mailing list