ICHEC
TECHNICAL REPORT
Getting Started :: First steps to using ICHEC resources
Getting Started, an introduction to ICHEC systems
Dr. Michael Browne ICHEC Computational Scientist
getting started
Introduction
Introduction Who can use the systems
ICHEC operates a number of High-Performance
Available hardware
Computing (HPC) systems for use by the Irish Research
Project classes
Community. This is a brief step-by-step guide to getting
Connecting to ICHEC systems
access to these systems and starting to run HPC tasks. It is
Available software
intended to simply address common questions and is not an
Batch job submission
exhaustive guide. If you still have questions or feel you have
Where can I find more help?
unusual requirements, by all means use the contact details
1 1 2 2 3 4 4 5
below to get in touch. being granted membership of class A, B & C projects, Ph.D.
Who can use the systems?
Students may also apply for their own Class C projects. This approach means a student can quickly get access to
Access to ICHEC systems is available free of charge to
significant HPC hardware and software without burdening
members of the Irish 3rd level research community subject to
their supervisors with administrative overhead. Those
an application process. The application process
outside Ireland can also use ICHEC systems if they are in
requirements depend on the level of resources requested.
active collaboration with Irish based researchers.
Access is centred around projects. Staff of 3rd level bodies apply to act as Principal Investigator (PI) of so called Class
While Class C projects are reviewed by ICHEC staff the
A,B or C projects. If their project application is successful
level of resources involved in a Class A or B application
they can then allow others to join their project and share the
means that a peer review process is merited. This process is
resources granted to them on ICHEC systems. In addition to
If you would like to speak in person with ICHEC staff please feel free to contact us via
[email protected], further details can be found on the following page: http://www.ichec.ie/contact_us
1
conducted by a Scientific Council which operates
operating system. As a result these systems are relatively
independently of ICHEC.
easy to use and an ideal place to start using HPC. When
ICHEC also collaborates with commercial and state
submitting a project application, time can be requested on
organisations in the field of HPC such as SGI and Met
either or both the Stokes and Stoney systems. Unless the
Éireann. Such collaborations are dealt with on a case-by-
RAM requirement per node is greater than 24GB, the time
case basis and are not the focus of this guide.
should be requested for Stokes. There are also two IBM Blue Gene systems,
Available Hardware
Schrödinger and Lanczos which are ideal for highly
ICHEC operates four primary systems. All use variants
scalable tasks. However, they are not generally not
of the Linux operating system. The most powerful and
suitable for those new to HPC. They are not available to
widely used system is called Stokes, a SGI Altix ICE
users of Class C projects. With this in mind this document
8200EX cluster with 3840 cores and 2 GB of RAM per
largely deals with the use of the Stokes and Stoney
core. Stoney is a Bull Bull Novascale R422-E2 cluster with
systems. Refer to ICHEC’s website or contact us for
512 cores and 6 GB of RAM per core. In many ways these
equivalent Blue Gene details.
are similar to conventional PCs running a Unix-like
More technical details of hardware on offer can be found here: http://www.ichec.ie/infrastructure
will be small research groups or individual
Project Classes
researchers. Successful applications are expected to
There are three different classes of project available:
lead to refereed publications.
•Class A - Class A projects are intended for consortia
•Class C - projects are intended to provide fast access
concerned with "Grand Challenge" problems. These
to modest resources with less review overhead. They
groups will require resources representing a
have multiple possible uses including: introductory
substantial fraction of ICHEC’s resources over a long
access for inexperienced HPC users; exploratory
period of time. Successful applications are expected to
access for researchers who need to develop, port,
yield high-impact scientific publications. Class A
optimise or benchmark codes; easier access for users
project project applicants are expected to have a
planning small scale runs with very modest
good knowledge of the characteristics of the code(s)
requirements; immediate access for researchers
which they intend to use - such as scalability properties
awaiting Class A or Class B application approval.
- before writing their proposal. For this reason, applicants who are not in such a position are advised
ICHEC is happy to engage with applicants and
to first apply for an exploratory Class C project in
endeavours to accept as many Class C applications as
order to undertake a basic scalability and
possible. The acceptance rate for peer reviewed Class B
performance study.
applications which follow our proposed guidelines is in excess of 90%. The following table summarises the
•Class B - projects are intended for the needs of the
properties of the three classes.
bulk of the research community. Typically applicants
Application process details can be found here: http://www.ichec.ie/services/full_national_service The application form can be found here: http://www.ichec.ie/project/apply
2
Class A
Class B
Class C
Max CPU core hours*
4500000
600000
25000
Max Storage
1,500 GB
500 GB
50 GB
Max Project Duration
36 months
24 months
12 months
Max Review Duration
12 weeks
6 weeks
2 weeks
* The times listed apply to Stokes, for Stoney these times should be divided 1.39
access to their network from which you may access ICHEC
Connecting to ICHEC Systems
systems.
Once you have been assigned a username and initial password you can connect to the systems using the Secure
It is of course possible to connect to ICHEC systems
Shell protocol (SSH). SSH is the most common method of
from MS Windows based systems also. However in order
remotely connecting to Unix-like systems for interactive
to so so you will have to install an SSH client. The most
access. If the machine from which you are connecting is
commonly used clients are putty and cygwin. Once SSH is
itself a Unix-like system e.g. an Apple Mac or a Linux
configured you can also use it for copying files to and from
system then it is very likely SSH is already installed. In
ICHEC’s systems using the scp or sftp commands which
which case from a terminal you can login in as follows:
most SSH packages will also provide. If you prefer to use a graphical means of exchanging files there are GUI packages such as Filezilla which support SSH connections.
$ssh
[email protected]
Running X Window based graphical applications on
If you wish to connect to an ICHEC machine other than
ICHEC’s systems is also possible. Again if you are using a
Stokes simply substitute the machine name. SSH encrypts
Unix-like system you probably already have a X server
all data which passes over the connection which greatly
installed. For security reasons it is necessary to pass or
benefits security. By default SSH uses network port 22. You
forward the X connection’s data over the SSH connection
will need to have outbound access on this port in order to
in the simplest case this can be accomplished by adding -X
successfully connect. Your local network security policy
to the command line as shown.
may mean that this is not possible by default. For instance you may first have to connect to another machine locally and then connect to ICHEC These settings are beyond
$ssh -X
[email protected]
ICHEC’s control and you should contact your local network administrator for more information. An indication that this
If you are connecting from a MS Windows machine
may be the case is if you get a "Connection timed out"
you will need to have Xming, Hummingbird Exceed or
error. If you believe the network port is open as discussed
similar installed and running on your workstation. You also
please submit an issue to the Helpdesk (see page 5).
need to ensure that X11 forwarding is enabled.
ICHEC accepts network connections only from the networks of its partner institutions. This means if you wish
Please note that while ICHEC is happy to assist with
to connect from home or abroad for example it is the
connection issues, which can sometimes be tricky to
responsibility of your home institution to provide for remote
configure in the first instance, the installation of any
3
required software on your system is ultimately your
typically run on fewer cores that as used during code
responsibly or that of your local IT support organisation.
development and testing.
However our use of standard, well proven and secure
•Interactive development jobs share the characteristics
methods minimises problems in the long run.
of normal development jobs but allow you to run command interactively on the compute nodes as you
Available software
might if there was no queuing system in place.
ICHEC centrally installs a large number of software
While sometimes cumbersome, batch processing
packages that are of common interest. Other packages are
allows you to exploit systems as large-scale resources. You
normally built on demand. You can get a list of the
do not have to wait for nodes to be available to run a job,
available packages using the following command.
you simply submit it and your job will be queued until the required resources are free and your job will then be run automatically. Batch processing allows the system to mix
$module avail
and match jobs of different sizes and durations to The module system is used to easily configure your
maximise the utilisation of the system. This makes it much
environment for the packages of interest to you, for
more economical to operate relative to an entirely
example prior to using the Intel compiler you should type
interactive scenario.
the following command. The job submission process works by the user producing a small script that details the resources required $module load intel-cc
by the job any environment setting it may need and the command required to actually run the job. This script is
Loading a module generally sets environment variables
then submitted to the scheduler, which reads it, checks
such as your PATH. You can list the modules you have
some basic parameters and enqueues the job for execution
loaded at any point by using the module list command.
as soon as possible. The following is a sample job submission script for the Stoney or Stokes systems. which gives an impression of the structure of such scripts.
$module list
#!/bin/bash #PBS -l nodes=4:ppn=12 #PBS -l walltime=1:00:00 #PBS -N my_job_name #PBS -A project_name #PBS -r n #PBS -j oe #PBS -m bea #PBS -M me@my_email.ie #PBS -V
It is important to include module load commands in your jobs submission scripts too. More details on using modules can be found here: http://www.ichec.ie/support/documentation/#env
Batch job submission To try to utilise compute resources in a fair and
cd $PBS_O_WORKDIR mpiexec ./my_program my_arguments
efficient manner, all compute jobs must be run through the batch queueing system. The system supports three main classes of jobs:
Note, the scheduler used on both Stokes and Stoney is
•Production jobs are normally day to day HPC jobs
called PBS, hence these scripts are often referred to as
that can run for long periods of time across a large
“PBS scripts”. The # symbol is required at the start of each
number of cores.
PBS directive. The line #PBS -l nodes=4:ppn=12 requests 48 cores in this case i.e. 4 nodes each of which have 12
•Development jobs are short jobs (