[PVFS2-developers] PVFS2 Request question
vilayann at mcs.anl.gov
Thu Jul 15 15:05:29 EDT 2004
So if I use it a sequence of PVFS_Request_contiguous(), it works as
expected and that is how I am currently using it in the writeback path
of the ctoo cache( though it is conceivably less efficient).
Also when the # of tuples is equal to 1 in the hindexed case, it works as
On Thu, 15 Jul 2004, Rob Ross wrote:
> This definitely looks like a bug to me.
> Does the problem occur if you use a contigous region for one or the other
> On Wed, 14 Jul 2004, Murali Vilayannur wrote:
> > Hi All,
> > I have a question on the usage/semantics of PVFS_Request_hindexed()
> > function. For comparison, I have attached an MPI test program
> > that uses MPI_Type_create_hindexed() & MPI_File_set_view() functions
> > to achieve the same thing.
> > what the reference test program does is allocates a chunk of memory,
> > randomly creates a
> > sorted set of tuples like <start location, size>, randomly selects a file
> > offset to write into, and then writes from the memory locations specified
> > through the set of tuples into a file. Note that I want the file's layout
> > to be similar to those generated in main memory, i.e using the same set
> > of tuples.
> > so essentially it tests non-contiguous writes both in memory and on disk.
> > it takes the following options
> > -f <file name>
> > -r <max number of tuples to generate. the final number that is generated
> > randomly is modulo this max. number>
> > -o <max offset in file. the final offset is modulo this max offset>
> > -b <buffer size in memory>
> > Suppose the ordered set of tuples are like so,
> > <o1,s1>....<o(i), s(i)>
> > and the offset is OF, then the file size should be OF + o(i) + s(i),
> > should n't it?
> > The above seems to be true when i execute the MPI test program. (both with
> > MPICH-1 and 2), but the pvfs2 test program does not work as desired.
> > Of course it is entirely possible that there is a bug in my code/usage.
> > Am I doing something wrong? Any suggestions/pointers welcome.
> > P.S: the pvfs2 test program does it right when r == 1, i.e when the # of
> > tuples to be written out is equal to 1 or when OF is equal to 0.
> > If you need me to dig up further, I would be glad to pitch in.
> > thanks,
> > Murali
More information about the PVFS2-developers