Class05 - MPI, CCR CompilingRunning PDF

MPI Quick Reference: Compiling/Running
M. D. Jones, Ph.D.
Center for Computational Research University at Buffalo State University of New York
High Performance Computing I, 2012
M. D. Jones, Ph.D. (CCR/UB)
HPC-I Fall 2012
1 / 24
Quickstart to Compiling & Running MPI Applications at CCR
Background
Background
This document covers the essentials of compiling and running MPI applications on the CCR platforms. It does not cover MPI programming itself, nor debugging, etc. (covered more thoroughly in separate presentations).
HPC-I Fall 2012
3 / 24
Modules
Modules Software Management System
There are a large number of available software packages on the CCR systems, particularly the Linux clusters. To help maintain this often confusing environment, the modules package is used to add and remove these packages from your default environment (many of the packages conict in terms of their names, libraries, etc., so the default is a minimally populated environment).
HPC-I Fall 2012
4 / 24
Modules
The module Command

module command syntax:
bash2.05b$ module h e l p Modules Release 3 . 1 . 6 ( C o p y r i g h t GNU GPL v2 1 9 9 1 ) : A v a i l a b l e Commands and Usage : + add | l o a d modulefile [ modulefile . . . ] + rm | unload modulefile [ modulefile . . . ] + s w i t c h | swap modulefile1 modulefile2 + d i s p l a y | show modulefile [ modulefile . . . ] + avail [ modulefile [ modulefile . . . ] ] + use [a | - append ] dir [ dir . . . ] + unuse dir [ dir . . . ] + update + purge + list + clear + help [ modulefile [ modulefile . . . ] ] + whatis [ modulefile [ modulefile . . . ] ] + apropos | keyword string + initadd modulefile [ modulefile . . . ] + initprepend modulefile [ modulefile . . . ] + initrm modulefile [ modulefile . . . ] + initswitch modulefile1 modulefile2 + initlist + initclear
HPC-I Fall 2012
5 / 24
Modules
Using module in Batch
If you change shells in your batch script you may need to explicitly load the modules environment: tcsh :
source $MODULESHOME/ i n i t / t c s h
bash :
. $ {MODULESHOME} / i n i t / bash
but generally you should not need to worry about this step (do a module list and if it works ok your environment should already be properly initialized).
HPC-I Fall 2012
6 / 24
Simple MPI Example
Objective: Construct a very elementary MPI program to do the usual Hello World problem, i.e. have each process print out its rank in the communicator.
HPC-I Fall 2012
8 / 24
Simple MPI Example
in C
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
# include < s t d i o . h> # include " mpi . h " i n t main ( i n t argc , char * * argv ) { i n t myid , nprocs ; i n t namelen , mpiv , mpisubv ; char processor_name [MPI_MAX_PROCESSOR_NAME ] ; M P I _ I n i t (& argc ,& argv ) ; MPI_Comm_size (MPI_COMM_WORLD,& nprocs ) ; MPI_Comm_rank (MPI_COMM_WORLD,& myid ) ; MPI_Get_processor_name ( processor_name ,& namelen ) ; p r i n t f ( " Process %d o f %d on %s \ n " , myid , nprocs , processor_name ) ; i f ( myid == 0 ) { MPI_Get_version (& mpiv ,& mpisubv ) ; p r i n t f ( " MPI V e r s i o n : %d.%d \ n " , mpiv , mpisubv ) ; } MPI_Finalize ( ) ; return 0; }
HPC-I Fall 2012
9 / 24
Simple MPI Example
U2: Intel MPI (Preferred!!)
U2: Intel MPI
There are several commercial implementations of MPI, Intel and HP currently being the most prominent (IBM, Sun, SGI, etc. all have their own variants, but usually are only supported on their own hardware). CCR has a license for Intel MPI, and it has some nice features: Support for multiple networks (Inniband, Myrinet, TCP/IP) Part of the ScaLAPACK support in the Intel MKL MPI-2 features (one-sided, dynamic tasks, I/O with parallel lesystems support) CPU pinning/process afnity options (extensive)
HPC-I Fall 2012
10 / 24
Simple MPI Example
Build the code with the appropriate wrappers:

[ u2 : ~ / d_mpisamples ] $ module l o a d i n t e l mpi i n t e l / 1 2 . 1 [ u2 : ~ / d_mpisamples ] $ module l i s t C u r r e n t l y Loaded M o d u l e f i l e s : 1) n u l l 2 ) modules 3 ) use . own 4 ) i n t e l mpi / 4 . 0 . 3 5) i n t e l /12.1 [ u2 : ~ / d_mpisamples ] $ m p i i c c o h e l l o . i m p i h e l l o . c # icc version [ u2 : ~ / d_mpisamples ] $ mpicc o h e l l o . i m p i . gcc h e l l o . c # gcc v e r s i o n
Unfortunately Intel MPI still lacks tight integration with PBS/Torque, and instead relies on daemons (launched by you) or the hydra task launcher to initiate MPI tasks.
HPC-I Fall 2012
11 / 24
Simple MPI Example
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
#PBS S / b i n / bash #PBS q debug #PBS l w a l l t i m e =00:10:00 #PBS l nodes =2:GM: ppn=2 #PBS jonesm@ccr . b u f f a l o . edu M #PBS e m #PBS N t e s t #PBS o subQ . o u t #PBS j oe # # Note t h e above d i r e c t i v e s can be commented o u t u s i n g an # a d d i t i o n a l " # " ( as i n t h e debug queue l i n e above ) # module l o a d i n t e l mpi i n t e l / 1 2 . 1 # # cd t o d i r e c t o r y from which j o b was s u b m i t t e d # cd $PBS_O_WORKDIR # # I n t e l MPI has no t i g h t i n t e g r a t i o n w i t h PBS, # so you have t o t e l l i t where t o run , b u t i t s mpirun # wrapper w i l l autod e t e c t PBS . # You can f i n d d e s c r i p t i o n o f a l l I n t e l MPI parameters i n t h e # I n t e l MPI Reference Manual . # See < i n t e l mpi i n s t a l l d i r > / doc / Reference_manual . p d f #
HPC-I Fall 2012
12 / 24
Simple MPI Example
27 28 29 30 31 32 33 34 35 36 37 38 39
export I_MPI_DEBUG=5 # n i c e debug l e v e l , s p i t s o u t u s e f u l i n f o NPROCS= ` cat $PBS_NODEFILE | wc l ` NODES= ` cat $PBS_NODEFILE | uniq ` NNODES= ` cat $PBS_NODEFILE | u n i q | wc l ` # # mpd based way : mpdboot n $NNODES f $PBS_NODEFILE v mpdtrace mpiexec np $NPROCS . / h e l l o . i m p i mpdallexit # # mpirun wrapper : mpirun np $NPROCS . / h e l l o . i m p i
HPC-I Fall 2012
13 / 24
Simple MPI Example
Intel MPI on Myrinet

The older U2 nodes have Myrinet - by default Intel MPI tries to run over the best available network:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 [ u2 : ~ / d_mpisamples ] $ c a t subQ . o u t Job 2949822. d15n41 . c c r . b u f f a l o . edu has requested 2 cores / p r o c e ss o r s per node . r u n n i n g m p d a l l e x i t on f09n35 LAUNCHED mpd on f09n35 v i a RUNNING : mpd on f09n35 LAUNCHED mpd on f09n34 v i a f09n35 RUNNING : mpd on f09n34 f09n35 f09n34 [ 0 ] DAPL s t a r t u p ( ) : t r y i n g t o open d e f a u l t DAPL p r o v i d e r from d a t r e g i s t r y : mx2 [ 1 ] DAPL s t a r t u p ( ) : t r y i n g t o open d e f a u l t DAPL p r o v i d e r from d a t r e g i s t r y : mx2 [ 2 ] DAPL s t a r t u p ( ) : t r y i n g t o open d e f a u l t DAPL p r o v i d e r from d a t r e g i s t r y : mx2 [ 3 ] DAPL s t a r t u p ( ) : t r y i n g t o open d e f a u l t DAPL p r o v i d e r from d a t r e g i s t r y : mx2 [ 1 ] MPI s t a r t u p ( ) : DAPL p r o v i d e r mx2 [ 0 ] MPI s t a r t u p ( ) : DAPL p r o v i d e r mx2 [ 3 ] MPI s t a r t u p ( ) : DAPL p r o v i d e r mx2 [ 2 ] MPI s t a r t u p ( ) : DAPL p r o v i d e r mx2 [ 0 ] MPI s t a r t u p ( ) : shm and d a p l data t r a n s f e r modes [ 1 ] MPI s t a r t u p ( ) : shm and d a p l data t r a n s f e r modes [ 3 ] MPI s t a r t u p ( ) : shm and d a p l data t r a n s f e r modes [ 2 ] MPI s t a r t u p ( ) : shm and d a p l data t r a n s f e r modes Process 1 o f 4 on f09n35 . c c r . b u f f a l o . edu Process 3 o f 4 on f09n34 . c c r . b u f f a l o . edu [ 0 ] MPI s t a r t u p ( ) : Rank Pid Node name Pin cpu [ 0 ] MPI s t a r t u p ( ) : 0 30112 f09n35 . c c r . b u f f a l o . edu 0
HPC-I Fall 2012
14 / 24
Simple MPI Example 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58
[ 0 ] MPI s t a r t u p ( ) : 1 30111 f09n35 . c c r . b u f f a l o . edu 1 [ 0 ] MPI s t a r t u p ( ) : 2 31983 f09n34 . c c r . b u f f a l o . edu 0 [ 0 ] MPI s t a r t u p ( ) : 3 31984 f09n34 . c c r . b u f f a l o . edu 1 [ 0 ] MPI s t a r t u p ( ) : I_MPI_DEBUG=5 [ 0 ] MPI s t a r t u p ( ) : I_MPI_FABRICS_LIST=dapl , t c p [ 0 ] MPI s t a r t u p ( ) : I_MPI_FALLBACK=enable [ 0 ] MPI s t a r t u p ( ) : I_MPI_PLATFORM=auto Process 0 o f 4 on f09n35 . c c r . b u f f a l o . edu MPI V e r s i o n : 2 . 1 Process 2 o f 4 on f09n34 . c c r . b u f f a l o . edu [ 1 ] DAPL s t a r t u p ( ) : t r y i n g t o open d e f a u l t DAPL p r o v i d e r from d a t r e g i s t r y [ 0 ] DAPL s t a r t u p ( ) : t r y i n g t o open d e f a u l t DAPL p r o v i d e r from d a t r e g i s t r y [ 3 ] DAPL s t a r t u p ( ) : t r y i n g t o open d e f a u l t DAPL p r o v i d e r from d a t r e g i s t r y [ 2 ] DAPL s t a r t u p ( ) : t r y i n g t o open d e f a u l t DAPL p r o v i d e r from d a t r e g i s t r y [ 0 ] MPI s t a r t u p ( ) : DAPL p r o v i d e r mx2 [ 1 ] MPI s t a r t u p ( ) : DAPL p r o v i d e r mx2 ... [ 3 ] MPI s t a r t u p ( ) : shm and d a p l data t r a n s f e r modes [ 0 ] MPI s t a r t u p ( ) : Rank Pid Node nameProcess 1 o f 4 on f09n35 . c c r . Pin cpu [ 0 ] MPI s t a r t u p ( ) : 0 30125 f09n35 . c c r . b u f f a l o . edu 0 Process 2 o f 4 on f09n34 . c c r . b u f f a l o . edu Process 3 o f 4 on f09n34 . c c r . b u f f a l o . edu [ 0 ] MPI s t a r t u p ( ) : 1 30126 f09n35 . c c r . b u f f a l o . edu 1 [ 0 ] MPI s t a r t u p ( ) : 2 32040 f09n34 . c c r . b u f f a l o . edu 0 [ 0 ] MPI s t a r t u p ( ) : 3 32041 f09n34 . c c r . b u f f a l o . edu 1 [ 0 ] MPI s t a r t u p ( ) : I_MPI_DEBUG=5 [ 0 ] MPI s t a r t u p ( ) : I_MPI_FABRICS_LIST=dapl , t c p [ 0 ] MPI s t a r t u p ( ) : I_MPI_FALLBACK=enable [ 0 ] MPI s t a r t u p ( ) : I_MPI_PIN_MAPPING=2:0 0 ,1 1 [ 0 ] MPI s t a r t u p ( ) : I_MPI_PLATFORM=auto Process 0 o f 4 on f09n35 . c c r . b u f f a l o . edu MPI V e r s i o n : 2 . 1
: : : :
mx2 mx2 mx2 mx2
b u f f a l o . edu
HPC-I Fall 2012
15 / 24
Simple MPI Example
Intel MPI on Inniband
The newest U2 nodes have Inniband (IB) as the optimal interconnect for message-passing, running an Intel MPI job should automatically nd and use IB on those machines (and they have 8 or 12 cores each, so adjust your script accordingly):
1 [ u2 : ~ / d_mpisamples ] $ c a t subQ . o u t
HPC-I Fall 2012
16 / 24
Simple MPI Example 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34
Job 2949999. d15n41 . c c r . b u f f a l o . edu has requested 12 cores / p r o c e ss o r s per node . r u n n i n g m p d a l l e x i t on k16n13a LAUNCHED mpd on k16n13a v i a RUNNING : mpd on k16n13a LAUNCHED mpd on k16n12b v i a k16n13a RUNNING : mpd on k16n12b k16n13a k16n12b [ 0 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 1 2 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 4 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 1 9 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 1 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 2 0 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 1 1 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 1 4 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 5 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 1 3 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 3 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 2 3 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 2 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 2 1 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 1 0 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 1 5 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 7 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 2 2 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 6 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 1 8 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 9 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 1 7 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 8 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes [ 1 6 ] MPI s t a r t u p ( ) : shm and t m i data t r a n s f e r modes
HPC-I Fall 2012
17 / 24
Simple MPI Example
35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
[ 0 ] MPI s t a r t u p ( ) : Rank Pid Node nameProcess 1 o f 24 on k16n13a . c c r . b u f f a l o . edu Process 2 o f 24 on k16n13a . c c r . b u f f a l o . edu Process 5 o f 24 on k16n13a . c c r . b u f f a l o . edu Process 11 o f 24 on k16n13a . c c r . b u f f a l o . edu ... Process 18 o f 24 on k16n12b . c c r . b u f f a l o . edu Process 20 o f 24 on k16n12b . c c r . b u f f a l o . edu Process 22 o f 24 on k16n12b . c c r . b u f f a l o . edu [ 0 ] MPI s t a r t u p ( ) : I_MPI_DEBUG=5 [ 0 ] MPI s t a r t u p ( ) : I_MPI_FABRICS_LIST= tmi , dapl , t c p [ 0 ] MPI s t a r t u p ( ) : I_MPI_FALLBACK=enable [ 0 ] MPI s t a r t u p ( ) : I_MPI_PIN_MAPPING=12:0 0 ,1 1 ,2 2 ,3 3 ,4 4 ,5 5 ,6 6 ,7 7 ,8 8 ,9 9 ,10 10 ,11 11 [ 0 ] MPI s t a r t u p ( ) : I_MPI_PLATFORM=auto Process 0 o f 24 on k16n13a . c c r . b u f f a l o . edu MPI V e r s i o n : 2 . 1
HPC-I Fall 2012
18 / 24
Simple MPI Example
Intel MPI on TCP/IP

You can force Intel MPI to run using TCP/IP (or a combination of tcp/ip and shared memory as in the example below) by setting the I_MPI_DEVICE variable, or equivalently I_MPI_FABRICS_LIST):
export I_MPI_DEBUG=5 export I_MPI_DEVICE=ssm # t c p / i p between nodes , shared memory w i t h i n
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Job 2950135. d15n41 . c c r . b u f f a l o . edu has requested 2 cores / p r o c e ss o r s per node . r u n n i n g m p d a l l e x i t on f09n35 LAUNCHED mpd on f09n35 v i a RUNNING : mpd on f09n35 LAUNCHED mpd on f09n34 v i a f09n35 RUNNING : mpd on f09n34 f09n35 f09n34 [ 3 ] MPI s t a r t u p ( ) : shared memory and s o c k e t data t r a n s f e r modes [ 0 ] MPI s t a r t u p ( ) : shared memory and s o c k e t data t r a n s f e r modes [ 2 ] MPI s t a r t u p ( ) : shared memory and s o c k e t data t r a n s f e r modes [ 1 ] MPI s t a r t u p ( ) : shared memory and s o c k e t data t r a n s f e r modes [ 1 ] MPI s t a r t u p ( ) : shm and t c p data t r a n s f e r modes [ 2 ] MPI s t a r t u p ( ) : shm and t c p data t r a n s f e r modes [ 3 ] MPI s t a r t u p ( ) : shm and t c p data t r a n s f e r modes [ 0 ] MPI s t a r t u p ( ) : shm and t c p data t r a n s f e r modes
HPC-I Fall 2012
19 / 24
Simple MPI Example
17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
Process 2 o f 4 on f09n34 . c c r . b u f f a l o . edu Process 1 o f 4 on f09n35 . c c r . b u f f a l o . edu Process 3 o f 4 on f09n34 . c c r . b u f f a l o . edu [ 0 ] MPI s t a r t u p ( ) : Rank Pid Node name [ 0 ] MPI s t a r t u p ( ) : 0 30385 f09n35 . c c r . b u f f a l o . edu [ 0 ] MPI s t a r t u p ( ) : 1 30384 f09n35 . c c r . b u f f a l o . edu [ 0 ] MPI s t a r t u p ( ) : 2 32261 f09n34 . c c r . b u f f a l o . edu [ 0 ] MPI s t a r t u p ( ) : 3 32260 f09n34 . c c r . b u f f a l o . edu [ 0 ] MPI s t a r t u p ( ) : I_MPI_DEBUG=5 [ 0 ] MPI s t a r t u p ( ) : I_MPI_DEVICE=ssm [ 0 ] MPI s t a r t u p ( ) : I_MPI_FABRICS_LIST=dapl , t c p [ 0 ] MPI s t a r t u p ( ) : I_MPI_FALLBACK=enable [ 0 ] MPI s t a r t u p ( ) : I_MPI_PLATFORM=auto [ 0 ] MPI s t a r t u p ( ) : MPICH_INTERFACE_HOSTNAME= 1 0 . 1 0 6 . 9 . 3 5 Process 0 o f 4 on f09n35 . c c r . b u f f a l o . edu MPI V e r s i o n : 2 . 1
Pin cpu 0 1 0 1
HPC-I Fall 2012
20 / 24
Simple MPI Example
Intel MPI Summary

Intel MPI has some real advantages: Multi-protocol support with the same build, by default gives you the "best" network, but also gives you the exibility to choose your protocol CPU/memory afnity settings Multiple compiler support (wrappers for GNU compilers, mpicc, mpicxx, mpif90, as well as Intel compilers, mpiicc, mpicpc, mpiifort) (Relatively) simple integration with Intel MKL, including ScaLAPACK Reference manual - on the CCR systems look at $INTEL_MPI/doc/Reference_Manual.pdf for a copy of the reference manual (after loading the module)
HPC-I Fall 2012
21 / 24
Simple MPI Example
Whither Goest Thou, MPI?
MPI processes - things to keep in mind: You can over-subscribe the processors if you want, but that is going to under-perform (but it is often useful for debugging). Note that batch queuing systems (like those in CCR) may not let you easily over-subscribe the number of available processors Better MPI implementations will give you more options for the placement of MPI tasks (often through so-called "afnity" options, either for CPU or memory) Typically want a 1-to-1 mapping of MPI processes with available processors (cores), but there are times when that may not be desirable
HPC-I Fall 2012
22 / 24
Simple MPI Example
Afnity
Intel MPI has options for associating MPI tasks to cores - better known as CPU-process afnity I_MPI_PIN, I_MPI_PIN_MODE, I_MPI_PIN_PROCESSOR_LIST, I_MPI_PIN_DOMAIN in the current version of Intel MPI (it never hurts to check the documentation for the version that you are using, these options have a tendency to change) Can specify core list on which to run MPI tasks, also domains of cores for hybrid MPI-OpenMP applications
HPC-I Fall 2012
23 / 24
Simple MPI Example
Summary
Summary - MPI at CCR
Use modules environment manager to choose your MPI avor I recommend Intel MPI on the clusters, unless you need access to the source code for the implementation itself. It has a lot of nice features and is quite exible. Be careful with task launching - use mpiexec whenever possible Ensure that your MPI processes end up where you want - use ps and top to check (also use MPI_Get_processor_name in your code). Also use the CCR ccrjobviz.pl job visualizer utility to quickly scan for expected task placement and performance issues.
HPC-I Fall 2012
24 / 24

Class05 - MPI, CCR CompilingRunning PDF

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Class05 - MPI, CCR CompilingRunning PDF

Transféré par

Droits d'auteur :

Formats disponibles

MPI Quick Reference: Compiling/Running

High Performance Computing I, 2012

M. D. Jones, Ph.D. (CCR/UB)

MPI Quick Reference: Compiling/Running

HPC-I Fall 2012

Quickstart to Compiling & Running MPI Applications at CCR

M. D. Jones, Ph.D. (CCR/UB)

MPI Quick Reference: Compiling/Running

HPC-I Fall 2012

Quickstart to Compiling & Running MPI Applications at CCR

Modules Software Management System

M. D. Jones, Ph.D. (CCR/UB)

MPI Quick Reference: Compiling/Running

HPC-I Fall 2012

Quickstart to Compiling & Running MPI Applications at CCR

The module Command

M. D. Jones, Ph.D. (CCR/UB)

MPI Quick Reference: Compiling/Running

HPC-I Fall 2012

Quickstart to Compiling & Running MPI Applications at CCR

Using module in Batch

M. D. Jones, Ph.D. (CCR/UB)

MPI Quick Reference: Compiling/Running

HPC-I Fall 2012

Simple MPI Example

M. D. Jones, Ph.D. (CCR/UB)

MPI Quick Reference: Compiling/Running

HPC-I Fall 2012

Simple MPI Example

M. D. Jones, Ph.D. (CCR/UB)

MPI Quick Reference: Compiling/Running

HPC-I Fall 2012

Simple MPI Example

U2: Intel MPI (Preferred!!)

U2: Intel MPI

M. D. Jones, Ph.D. (CCR/UB)

MPI Quick Reference: Compiling/Running

HPC-I Fall 2012

Simple MPI Example

U2: Intel MPI (Preferred!!)

Build the code with the appropriate wrappers:

M. D. Jones, Ph.D. (CCR/UB)

MPI Quick Reference: Compiling/Running

HPC-I Fall 2012

Simple MPI Example

U2: Intel MPI (Preferred!!)

M. D. Jones, Ph.D. (CCR/UB)

MPI Quick Reference: Compiling/Running

HPC-I Fall 2012

Simple MPI Example

U2: Intel MPI (Preferred!!)

M. D. Jones, Ph.D. (CCR/UB)

MPI Quick Reference: Compiling/Running

HPC-I Fall 2012

Simple MPI Example

U2: Intel MPI (Preferred!!)

Intel MPI on Myrinet

M. D. Jones, Ph.D. (CCR/UB)

MPI Quick Reference: Compiling/Running

HPC-I Fall 2012

Simple MPI Example 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58

U2: Intel MPI (Preferred!!)

mx2 mx2 mx2 mx2