Hi Murali,
> Aren't the two issues (adding a new parameter to the system interface and
> the ability to create dfiles on specific nodes) sufficiently
> orthogonal/different?
Yes they are, but if we extend the system interface then we could easily add 
another parameter if necessary and pending changes to the system interfaces 
can be done in one modification. Thats why I was asking.
> Can you tell us what exactly was the problem with the new distribution?
I was thinking that all distributions could benefit from such a mechanism. 
Maybe the metatdata server for that file could be selected on the same way, 
too ?
Within the distribution itself the server names are not available I guess, so 
there has to be some hack code outside. Also, I think due to dynamic 
migration etc. it is no good idea to write a distribution which fixated the 
datafiles. 
> Adding server locations to sys_create() does not seem like a great way
> forward.
> Perhaps, we could use the extended attribute features on a directory
> to indicate locations of servers for dfiles on a file create? Look at
> this..
This might be possible and is a another cool way, however we want the user to 
set it for example with MPI_hints before the file is opened and only to set 
it for a single file. For example somehow to use dedicated servers for this 
file of a single file system, this might be used to place the datafiles on 
the same nodes if the client also acts as a server or to place the 
datafiles "close" to the client on a dataserver. It seems to be rather 
inefficent to change the extended attributes of a directory then create the 
file and change it back.

Thanks for the fast response,
enjoy playing ultimate :)
Julian
_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers

Reply via email to