On Oct 31, 2017, at 05:46, Mohr Jr, Richard Frank (Rick Mohr) <[email protected]> 
wrote:
> 
>> On Oct 30, 2017, at 4:46 PM, Brian Andrus <[email protected]> wrote:
>> 
>> Someone please correct me if I am wrong, but that seems a bit large of an 
>> MDT. Of course drives these days are pretty good sized, so the extra is 
>> probably very inexpensive.
> 
> That probably depends on what the primary usage will be.  If the applications 
> create lots of small files (like some biomed programs), then a larger MDT 
> would result in more inodes allowing more Lustre files to be created.

With mirroring the MDT ends up as ~2.4TB (about 1.2B files for ldiskfs, 600M 
files for ZFS), which gives an minimum average file size of 120TB/1.2B = 100KB 
on the OSTs (200KB for ZFS).  That said, by default you won't be able to create 
so many files on the OSTs unless you reduce the inode ratio for ldiskfs at 
format time, or use ZFS (which doesn't have a fixed inode count, but uses twice 
as much space per inode ob the MDT). 

Having a larger MDT isn't bad if you plan future expansion.  That said, you 
would get better performance over FDR if you used SSDs for the MDT rather than 
HDDs (if you aren't already planning this), and for a single OSS you probably 
don't need the extra MDT capacity.  With both ldiskfs+LVM and ZFS you can also 
expand the MDT size in the future if you need more capacity.

Cheers, Andreas
--
Andreas Dilger
Lustre Principal Architect
Intel Corporation







_______________________________________________
lustre-discuss mailing list
[email protected]
http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org

Reply via email to