Thanks for the response. Just to clarify, lustre 2.1 mdt uses a 128bit
file identifiers (FID), while 1.8 uses 64bit FID. However, this
difference is transparent to the 1.8 clients ?
The reason I ask because when I start a 1.8 client against a 2.1 MDT, I
receive this message:
Lustre: Server scratch-MDT0000_UUID version (188.8.131.52) is much newer
than client version (1.8.9)
Lustre: client supports 64-bits dir hash/offset!
I previously upgraded from 1.4 -> 1.6 and 1.6-> 1.8. The 1.8 -> 2.1
upgrade was much smoother (we had vendor support this time around).
Kudos for all the progress removing upgrade bumps.
On 02/06/2014 09:11 PM, Dilger, Andreas wrote:
The fid-in-dirent feature is only relevant on the MDT. It does not
directly affect the client or OST, but will affect the performance
that the client sees when accessing files created under 2.x.
There is a noticeable performance improvement (or more specifically
no performance degradation) for 2.x files with this feature enabled.
The main reason it is not enabled by default is that it impacts the
ability to downgrade to 1.8 again safely. Once you know you don't
want to downgrade again it is ok to enable it.
> On Feb 6, 2014, at 13:43, "Chris Hunter" <chris.hunter(a)yale.edu>
> We are in process of upgrading a lustre FS. We currently have the
> MDS and OSS servers running 2.1.5 (upgraded from 1.8.9). However
> our client are still running 1.8.9.
> I am trying to clarify if the "FID-in-dirent" feature (ie. tunefs
> -O dirdata) is backwards compatible with 1.8.9 clients ?
> I have seen a few talks (SC11?) that says it should be. However the
> 2.x lustre operations manual at hpdd-intel doesn't *explicitly* say
> it will work with older clients.
> Is there high risk involved enabling "dirdata" when running old
> 1.8.9 clients ?
> Any opinions if there is benefit in using this flag ?
> thanks, chris hunter
> _______________________________________________ HPDD-discuss
> mailing list HPDD-discuss(a)lists.01.org