pptx - Mississippi Center for Supercomputing Research - University ...

compliantprotectiveΛογισμικό & κατασκευή λογ/κού

1 Δεκ 2013 (πριν από 3 χρόνια και 8 μήνες)

68 εμφανίσεις

What is MCSR?


Who is MCSR?

What Does MCSR Do?

Who Does MCSR Serve?

What Kinds of Accounts?

Why Does Mississippi Need Supercomputers?

What Kinds of Research?

What Kinds of Instruction?

What Kinds of Workshops?

How Much Does it Cost?

What Kinds of Software?

What Supercomputers and Clusters?



What is MCSR?


Mississippi Center for Supercomputer Research


Established in 1987 by the Mississippi Legislature



Mission:

Enhance Computational Research Climate


at Mississippi’s 8 Public Universities




also:

Support High Performance Computing


(HPC) Education in Mississippi

What Does MCSR Do?


We make Mississippi
scientists
:


-

more
competitive

for federal grants


-

more
productive

in research


We provide
extraordinary learning opportunities
for
Mississippi college
students


-

instructional
accounts


-

computing
workshops


-

helpdesk
support




Who Does MCSR Serve?


MCSR serves faculty, researchers, and students at all
of Mississippi’s 8 public universities



Alcorn State University



Jackson State University



Mississippi Sate University



Mississippi Valley State University


The University of Southern Mississippi


Delta State University


The University of Mississippi


Mississippi University for Women



Who Uses More MCSR Computing Time?

MSU, 38,771,
4%

USM, 25,279,
3%

UM, 172,946,
19%

JSU, 663,177,
74%

Other, 2,145,
0%

CPU Hours Used by IHL

All MCSR Systems

FY 2009 (1st QTR)

2%

1%

10
%

87
%

SWEETGUM


1%

28%

71%

REDWOOD

7%

5%

16%

71%

MIMOSA

What Types of Computing Access is Available?


Research Accounts


-

provided for faculty and student “researchers”


-

good for the
duration of employment or enrollment


Instructional Accounts


-

provided at an instructor’s request


-

for all students enrolled in a semester course


-

valid for the
duration of the semester


-

can be “converted” to research account



Research vs. Class Accounts?




New Class Accounts,
266, 73%

New Research
Accounts, 100, 27%


MCSR Accounts Created FY 08

New Class Accounts
New Research Accounts
Who Uses Research Accounts?

usm

6%

msu

5%

UM

41%

jsu

46%

other

2%

MCSR Research Accounts


by IHL

September 2008

usm
msu
um
jsu
other
Who Uses Instructional Accounts?

UM, 2

DSU, 8

USM, 1

Courses Supported with MCSR Class
Accounts, FY 2008

UM
DSU
USM
UM, 5

DSU, 8

USM, 1

Courses Supported with MCSR Class
Accounts, FY 2007

UM
DSU
USM
UM, 7

DSU, 5

USM, 1

Courses Supported with MCSR Class
Accounts, FY 2006

UM
DSU
USM
MVSU, 2

UM, 13

DSU, 2

USM, 1

Courses Supported with MCSR Class
Accounts, FY 2005

MVSU
UM
DSU
USM
What Types of Courses Use MCSR?


Over 82 University Courses Supported since 2000




C/C++, Fortran,

MPI, OpenMP,

MySQL, HTML,

Javascript,

Matlab, PHP,

Perl, ….






http://www.mcsr.olemiss.edu/education.php


Why Do Mississippi Researchers
Need Supercomputers?


Economics: researchers in a poor state like Mississippi can
still make a big splash.


Computational simulations are faster, cheaper, and safer
than laboratory experiments alone.


What Kinds of Research @ MCSR?

90% of MCSR calculations are computational chemistry


Cleanup of high explosive materials
.

Design of high energy density rocket fuels

The chemical underpinnings of high powered lasers

Mutation studies of enzyme activity

Designing weather
-
proofing coatings for machinery


Other Areas


Hurricane forecasting


Blast resistant Coatings


Better 3
-
D imaging for diagnosing brain tumors


What Types of Workshops by MCSR?



MCSR consultants taught over 140 free seminars in
FY08.


Over 60 training topics available, and growing.


Fixed schedule or on
-
demand.


Unix/programming, Math Software, Stats Software,
Computational Chemistry Software

Do Researchers and Students Pay to Use MCSR?


No.
MCSR services are provided at no cost to the
individual, department, or institution.


Funded researchers may ask for priority access
.


Mississippi researchers may claim the value of MCSR
computing services received as an in
-
kind contribution
from their institution when seeking federal grants.



How Much Does MCSR Cost Mississippi?

$-
$500,000
$1,000,000
$1,500,000
$2,000,000
$2,500,000
$3,000,000
FY88
FY89
FY90
FY91
FY92
FY93
FY94
FY95
FY96
FY97
FY98
FY99
FY00
FY01
FY02
FY03
FY04
FY05
FY06
FY07
FY08
FY09
MCSR Budget by Fiscal Year

Original
Revised
Special Funds
What Is the Value of MCSR to Mississippi?


RETURN ON INVESTMENT
=





GRANTS SUPPORTED

/
MCSR BUDGET


For FY 2008:







$32,832,097

/

$845,535


= $38.30 per dollar spent

What Software Environments @
MCSR


Programming




C/C++, FORTRAN, Java, Perl, PHP, MPI…


Science/Engineering



PV
-
Wave, IMSL, GSL, Math Libraries,
Abaqus


Math/Statistics



SAS, SPSS,
Matlab
, Mathematica


Chemistry



Gaussian, Amber,
NWChem
, GAMESS,


CPMD, MPQC,
MolPro
, GROMACS

What Supercomputers @ MCSR?

Supercomputers at MCSR:
sweetgum

-

SGI Origin 2800 128
-
CPU Supercomputer

-

64 GB of shared memory






Supercomputers at MCSR:
redwood


-

224 CPU SGI
Altix

3700 Supercomputer

-

224 GB of shared memory






Supercomputers at MCSR:
mimosa

-
253 CPU Intel Linux Cluster


Pentium 4

-
Distributed memory


500MB


1GB per node

-
Gigabit Ethernet






Supercomputers at MCSR:
mimosa

Supercomputers at MCSR:
sequoia

-
22 nodes

-
176 cores

-
352 GB Memory

-
20 TB Storage

-
InfiniBand

Interconnect






Supercomputers at MCSR:
sequoia

Introduction to

Parallel Programming at MCSR


Message Passing Computing


Processes communicate via calls to message passing library routines


Programmers “parallelize” algorithm and add message calls


At MCSR, this is via MPI programming with C or Fortran


Sweetgum



Origin 2800 Supercomputer (128 CPUs)


Mimosa


Beowulf Cluster with 253 Nodes


Redwood


Altix

3700 Supercomputer (224 CPUs)


Sequoia


Altix

XE 310
InfiniBand

Cluster (176 cores)


Shared Memory Computing


Threads coordinate/communicate results via shared memory variables


Care must be taken not to modify the wrong memory areas


At MCSR, this is via
OpenMP

programming with C or Fortran on
sweetgum
, redwood, or sequoia


Speed
-
Up

http://www.mcsr.olemiss.edu/Engr692_TimingWorshkeet.xls

What MCSR Systems for

USM Class Accounts

Sweetgum


MPI or
OpenMP


1 to 16 CPUs


Up to 900mb per CPU


PBS scripts preferred




#PBS

l
ncpus
=4


Interactive computations will be killed after 30 minutes


Queues: SM
-
4P, SM
-
8P, MM
-
8P, MM
-
16P


Processors: Mix of 195 MHz and 300 MHz


O/S:
Irix

(like Unix)


Compilers: SGI’s Fortran, C/C++, GNU C/C++, w/ SGI MPT



What MCSR Systems for

USM Class Accounts

Mimosa


MPI


1 to 18 nodes


400 GB Memory per node


PBS Scripts Only (no interactive jobs allowed)



#PBS

l nodes=4


Queues: MCSR
-
CA


Processors: single 1.4 GHz P4/node


O/S: SUSE Linux 10.3


Compilers: Portland Group (PGI) Fortran, C/C++ w/
mpich


qstat


f (to find out what nodes your job is running on)








What MCSR Systems for USM Class Accounts

Sequoia


OpenMP

(multiple processors on the same node)


MPI (multiple processors on the same or different nodes)


Hybrid (
OpenMP

within node, MPI across nodes)


1 to 4 nodes, 1 to 8 CPUs per node


PBS Queues: SM
-
4P (for up to 4 CPUs on 1 node)





MCSR
-
Test (up to 8 CPUs on each of 4 nodes)


PBS Scripts Only (no interactive jobs allowed)


#PBS

l nodes=4:ppn=8 (to run on all 8 CPUs of all 4 nodes)


#PBS

l
ncpus
=8:select=



16 GB Memory per node (2 GB per CPU)


qstat


f (to find out what nodes your job is running on)










Sequoia for USM Class Accounts

To run on 4 nodes, and 8 processors per node (32 processes)


#PBS

l nodes=4:ppn=8

To run on 2 nodes, and 4 processors per node (8 processes)


#PBS

l nodes=2:ppn=4

To run on 1 node, and up to 8 processors (
OpenMP
)


#PBS

l nodes=1:ppn=8

To run on 8 processors, regardless of number of nodes


#PBS

l
ncpus
=8

To run 8 processors, with preferences about node placement


#PBS

l
ncpus
=8


#PBS

l place=scatter (distribute across as many nodes as can)


#PBS

l place=pack (pack processes onto as few nodes as can)


#PBS

l place=free (place processes on first available processors)










Parallel Efficiency