Robert Latham wrote:
On Thu, Mar 02, 2006 at 12:01:15PM +0100, Phil Carns wrote:
create-round-robin.patch:
--------------------------
Wonder if balancing out the load across the servers more evenly will
help us out with metadata-intensive workloads? Well, it's been
comitted.
We haven't tested to see what the result is at the pvfs2 level, but we
can see the difference in balance when monitoring the throughput at the
storage device level.
makefile-mpi-objects.patch:
---------------------------
Small cleanup to the makefile for the test subdirectory. It previously
linked all MPI_MISC objects into every MPI test program, whether needed
or not. This patch adjust things so that the misc. objects are only
linked into relevant executables.
I'm going to defer this one a bit: slang's added some mpi tests i
think so we'll have to account for those new ones.
Sounds good. I have some half-completed mpi test programs that will
eventually require this patch to avoid symbol collisions, but it looks
like it is going to be a while before I finish those up and pass them along.
sys-getattr-mask.patch:
---------------------------
This fixes a few problems in the handling of the datafile_count and
dirent_count fields in sys_getattr. They were being left out of some of
the object attribute <--> system attribute conversion functions. The
dirent_count was also not being copied out to the caller upon completion.
good catch. was this causing bugs for you or did you just notice it
while walking through the source code? comitted this one too.
This was causing a bug for us, but we were using the system interface
getattr function directly in a user space program to see it. I'm not
sure how/if it gets manifested in the VFS or MPI/IO path.
-Phil
_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
