The answer I sent you talks about the distribution of directory entries
themselves, rather than the metadata (/inode) information about each file.
The experimental release behaves exactly the same as other releases with
respect to these handles. The metadata storage for files will continue to be
selected from all available meta handles. This distribution is randomly
spread in order to minimize bottlenecks at metadata servers. Using
readdir_plus rather readdir will help minimize the cost of consolidating
inode information when listing directories.

 

Sorry for the confusion.

Elaine

 

From: Elaine Quarles [mailto:[email protected]] 
Sent: Tuesday, September 20, 2011 12:27 PM
To: '欧佳欣'; 'pvfs2-developers'
Subject: RE: [Pvfs2-developers] Some quesions about the create operation in
pvfs2

 

That was just a simplification made in the early version. The
soon-to-be-released version will store directory entries all on a single
metadata server for directories containing a small number of directory
entries (where the definition of "small" is configurable). When the number
of entries grows beyond the threshold, the entries will be distributed
across multiple metadata servers.

 

Hope this helps,

Elaine

 

From: [email protected]
[mailto:[email protected]] On Behalf Of ???
Sent: Tuesday, September 20, 2011 12:12 PM
To: pvfs2-developers
Subject: [Pvfs2-developers] Some quesions about the create operation in
pvfs2

 

       In orangefs-2.8.3-EXP, when we create a file , we will call the
function PINT_cached_config_get_meta_handle , and it just randomly select
the server to store the meta data regardless of the size of its parent
directory.So we can conclude that the inode of a file in a directory may
store in different Metadata servers , even though  the directory contains a
few files.

       So I want to know why should pvfs2 use this method to select a
metadata server to store the inode?

I just  know that there is no need to use this method when the directory is
very small, in this case , it may even reduce the system performance .

_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers

Reply via email to