Introduction to ACENET Accelerating Discovery with Computational Research May, 2015

Introduction to ACENET

Accelerating Discovery with Computational Research May, 2015

What is ACENET?

Shared regional resource for...

…

high-performance computing (HPC)

… remote collaboration

… visualization

Mandate

We facilitate

research

,

education

and

innovation

in

Atlantic Canada by providing Advanced Research

Computing resources and expertise.

History

Began in 2003.

CFI funding $25 million – 2006. Similar organizations developed in other regions.

The four regions have come together with a national umbrella organization called Compute Canada.

Compute Canada operates as a federated model; ACENET is Compute Canada’s regional partner in Atlantic Canada.

Compute Canada helps to:

•  better allocate scarce

funding dollars;

•  provide researchers with

seamless access to

infrastructure and resources nationally; and

•  enable the advanced

computing community to have a single, strong voice in the national forum.

Computing Resources Available

Big Data and Data Analytics tools and environments

Leading edge GPU computing systems

Highspeed, secure file transfer with the Globus Portal

Extensive software library

GenAP computing platform to access and use genomic datasets

Compute Canada’s Cloud computing and environment

Data storage and back-up systems that provide stability and security options over your desktop

Desktop and mobile videoconferencing

Collaboration rooms at all ACENET partner institutions to deliver and view large videoconferences.

ACENET Expertise Available

Determining computing resources needed

Designing, optimizing and

troubleshooting computer code

Customizing tools

In-depth collaboration where needed

Group and individual training from novice to advanced

Accessible and responsive support staff

Installing, managing and

maintaining advanced research computing equipment

Access to Digital Humanities expertise

Access to 3D Visualization expertise

What is

Advanced

Research

High Performance Computing (HPC) Data Mining Computationally Intensive Data Intensive Parallel Computing High Performance Data Analysis (HPDA) Data Intensive Simulation & Analysis Supercomputing

Cloud

BIG

DATA

Answer:

Any computing that is too big for a

desktop.

What is High Performance Computing

Many CPUs working on one problem

• 

Many “serial” jobs running at once; or

• 

Many CPUs working in close coordination.

• 

Communication between CPUs is key!

Advantages

• 

Faster results

• 

Larger problems or more detailed simulations

ACENET Computers

Four clusters

7000 CPU cores

20 TB RAM

500 TB disk storage

Plus tape

Plus software

Remember: Shared resources!

Hundreds of users

aggregated

numbers

What is a Cluster?

scheduler

"Grid Engine"

How Do I Get an Account?

Go to

http://www.acceleratediscovery.ca/get-an-account/

No, you don’t have to pay any money.

Just tell us once a year about the great research you’re

doing so the taxpayers can be convinced to continue

funding this resource.

How Do I Get Connected?

Logging in:

ssh Secure Shell

Moving Data Around:

sftp Secure File Transfer Protocol

Scp Secure Copy

Both provided with Mac OS X or Linux

ssh -X username@mahone.ace-net.ca sftp username@placentia.ace-net.ca

Windows?

Try MobaXterm or PuTTY and WinSCP. Just google for them

Applications

http://www.acceleratediscovery.ca/wiki/Software

Software requests to:

support@ace-net.ca

Molecular Gaussian Gromacs Q-Chem VASP NAMD WebMO VMD ... Bio AbySS Bowtie rapsearch2 MrBayes Migrate-n PhyML PLINK ... Earth &Ocean CDO NCO GMT ... Engineering ANSYS Fluent CFX OpenFOAM … Math&CS Sage GAP DiVinE ... Astro StarLab IRAF ...

Software Development

http://www.acceleratediscovery.ca/wiki/Software

Data Analysis

http://www.acceleratediscovery.ca/wiki/Software

Python

R

Octave

Matlab*

*Distributed Computing Server

Available Compilers At ACENET

C, C++, and Fortran

Portland Group (PGI) version 8

•  pgcc, pgCC, pgf77, pgf90, pgf95

GNU suite versions 3 and 4

•  gcc, g++, gcc4, g++4, gfortran, g77

Intel Compiler Suite XE 2011 •  icc, icpc, ifort

SunStudio version 12

•  cc, CC, f77, f90, f95

Environment Modules

$ module list

Currently Loaded Modulefiles:

1) pgi/8.0-6 3) totalview/8.8.0-2

2) openmpi/pgi/1.2.9

$ gcc –-version

gcc (GCC) 3.4.6 ...

$ module avail gcc

gcc/4.4.3(default) gcc/4.6.4 gcc/4.8.0

$ module load gcc

$ gcc --version

gcc (GCC) 4.4.3 ... ←

changed!

$ module unload gcc

MPI

OpenMPI is default implementation

Use modules to vary the underlying compiler, e.g.

$ module unload openmpi

$ module load gcc openmpi/gcc

Parallel Job Processing

Notice that process count and host list is passed automatically from Grid Engine to Open MPI.

$ cat jobscript

#$ -cwd

#$ -j yes

#$ -l h_rt=48:00:00

#$ -l h_vmem=2G

#$ -pe ompi* 32

module purge

module load gcc openmpi/gcc

mpirun ./application

Where To Go For Help

Online:

ACEnet wiki

Email support:

•  Please supply as much info as possible

What Cluster Should I Use?

It doesn't matter,

ONE ACCOUNT IS GOOD EVERYWHERE

Multiple serial jobs?

Shared memory parallel?

MPI?

Terabytes of storage?

Some software has only

local license

Fit in anywhere

Fundy best; Glooscap good

Mahone, Placentia best; Fundy good Probably Glooscap

e.g. Matlab DCS & Gaussian only at Placentia

See www.acceleratediscovery.ca/wiki/Software for availability, or contact your Computational Research Consultant