Yan,
your speedups for 2 processors look about right. Two MPI processes on a
two-processor machine will not run as fast as a true multithreaded
application on a two-processor machine. A 61% speedup is pretty good.
Remember that MPI is operating as though everything is out-of-process all
the time, even when they're both on the same machine. Your speedup on 3
processors is very good, you added 50% more power and almost got a 50%
speedup, 1.6 x 1.5 = 2.4, and you got 2.31. For 4 processors, it looks like
you've reached a communication limit. I think what has happened is that
your problem on 4 processors requires more MPI communication compared to the
processor work and so you're not getting any more speedup. As you
distribute a given amount of work across multiple processors, you eventually
always reach a point where you don't get faster as you add more processors
because the communication time takes longer than the computation time.
Damien
----- Original Message -----
From: "choy hau yan" <choyhauyan_at_[hidden]>
To: <lam_at_[hidden]>
Sent: Monday, February 23, 2004 8:18 PM
Subject: LAM: distributed shared memory ?
> hi,I have ask this question before but still confuse,
> so..hope that someoen can give more detail
> answer.thanks.
>
> Q1:
> I got 2 processors that in shared memory and then
> connect with TCP/IP to another 2 processors in shared
> memory.
>
> I use mpisend/recv for communication, but why can't I
> call this shared distributed memory?
> The speedup with this architecture is very low.why?
>
> speedup:
> 2 processor: 1.61
> 3 processor: 2.31
> 4 processor: 2.30
> actually with shared memory, the speedup is more high
> that distributed becasue almost no cos communication
> in shared memory.hope that some one can answer my
> question. thanks..
>
> Q2,
> I am using a big file of fortran,is it more better to
> compile my file with makefile than mpicc ?
> Is this the main cause that my speedup is low?
>
> thanks you...
>
>
> __________________________________
> Do you Yahoo!?
> Yahoo! Mail SpamGuard - Read only the mail you want.
> http://antispam.yahoo.com/tools
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
|
