[PVFS-developers] why do lookup after create/mkdir?

[Rob Ross]

Hi,

You are right that we could combine returning statistics with those calls.  
It would require changing the wire protocol somewhat, but it would be 
faster.

To really answer your question though, we're spending our development time 
on PVFS2.  Any change that we make to PVFS1 opens up the possibility of 
more bugs being introduced, and we really would prefer that PVFS1 be in 
more of a maintenance mode.  However, with all the great patches people 
are providing, it continues to grow regardless.

Were you to provide a patch, we would certainly consider it.  I would 
suggest first creating a patch that added the stat returns to the create 
and mkdir calls, handling the EEXIST case by returning the stats for the 
existing entity (and still returning an error).  The kernel-side patch 
would be second, and would need to address both the pvfsd and kpvfsd as 
necessary.

Regards,

Rob

On Thu, 25 Dec 2003 songdm@lenovo.com wrote:

> Hi,all,
> 
> I find that in pvfs_file_create function, ll_pvfs_lookup is called after
> ll_pvfs_create. And  pvfs_mkdir do the same after ll_pvfs_mkdir. It results
> in once more communication between client node and mgr node. I think it
> isn't neccessory. Why not do all the tasks in ll_pvfs_create? Thus we can
> improve throughput of pvfs.
> 
> Best regards
> Song Dongmei