thanks for answer.
for question 1,
can we call distributed memory? because it also got 2
processors share the same memory but they also
connecting with another CPU.
quwstion 2,
If master sending small value array,all the data is
correct. if sending big array, it got problem. Is that
posible use some command do the clear off in buffer
before use another mpi_send.(I used many mpi_send in
one time).
Is that posible some data lost or change when sending
big data form one processors to another ? this happen
because of data lost in network wire while
transfering.
thanks.
--- Jeff Squyres <jsquyres_at_[hidden]> wrote:
> On Mon, 28 Jul 2003, choy hau yan wrote:
>
> > I am using 4 processor to run a fortran program.(2
> processor in a CPU.)
> > two CPU connect with network.
> >
> > question 1:
> >
> > can I know what the name of this architecture. It
> called shared
> > distributed memory?
>
> Sounds more like distributed memory. With MPI,
> you're using multiple
> processes, each with its own unique memory space.
> Hence, it's distributed
> memory.
>
> > question 2:
> >
> > I am sending a file let say PLK2(400,12,12).
> > call
> mpi_send(PLK2,57600,mpi_real,0,ja,mpi_comm_world,
> ierr)
> >
> > I just sending a file from master to slave to run
> then
> > send back to master.
> > [snipped]
> > When starting the calculation, all the answer from
> salve is same, but
> > after some do loop, I found that the correct
> answer from slave when send
> > to master become not correct.
> >
> > For example 2-5 value in this array :
> PLK2(400,12,12) is not the same.
> > the value is correct in slave but after receive
> from master, the number
> > become different.
>
> It's hard to tell from your description, but it
> sounds like you've got
> some kind of buffer mismatch in there somewhere.
>
> Try filling your buffers with obvious sentinel
> values (e.g., have the
> master send all 17's or something like that, and
> ensure that the slave(s)
> receive it fine. Then have the slaves send back a
> buffer filled with
> their rank number to the master and see if the
> master receives it
> properly. And so on.
>
> See if this helps find the error.
>
> --
> {+} Jeff Squyres
> {+} jsquyres_at_[hidden]
> {+} http://www.lam-mpi.org/
> _______________________________________________
> This list is archived at
http://www.lam-mpi.org/MailArchives/lam/
__________________________________
Do you Yahoo!?
Yahoo! SiteBuilder - Free, easy-to-use web site design software
http://sitebuilder.yahoo.com
|
