Le Thu, 4 Sep 2008 15:25:06 -0400 vous écriviez:

>   Quick question that I can not seem to find an answer.  Can I add
> new IO server to an
> existing PVFS2 filesystem without reformatting the entire 
> filesystem/corrupting the
> existing data?  Also, will it over time balance the load over the
> entire filesystem?

I've made some tests and I'm currently doing some more tests increasing
a PVFS2 storage cluster size (using 2.7.1). I've found that adding
another node to an existing cluster works aarently fine, but for
_one_file_ : the 4th file written (no other) gets corrupted and is
unavailable. However, if you create 4 files, remove the 4th, then go on
writing, then increase the cluster, apparently all goes well and dandy.

I've made my first test with different storage size on the nodes; I'm
about to re-test it with nodes of all the same size (39.7 TB per node,
6 nodes, total ~ 240 TB ) to see if there's still this bug.

You can quite easily redispatch existing files on the cluster with
a  defrgmentation script, many can be found for Linux on the net. 
However I plan to write one using pvfs2-cp to be more efficient,
something like 

find . -type f -exec pvfs2defrag {} \;

With pvfs2defrag roughly as follow (NOT TESTED):

#!/bin/sh

SOURCEFILE=$1
DESTFILE="."$1".defrag.$$"

pvfs2-cp $SOURCEFILE $DESTFILE
mv $SOURCEFILE $SOURCEFILE.bak
mv $DESTFILE $SOURCEFILE
pvfs2-rm $SOURCEFILE.bak


Of course this is not enough; we should get sure no process uses the
file before attempting to defrag it, make some sanity checks on the
file after the copy, etc.

-- 
--------------------------------------------------
Emmanuel Florac               www.intellique.com   
--------------------------------------------------

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

Reply via email to