16 page newsletter - Trinity Centre for High Performance Computing · 2017-12-12 · interconnect...
Transcript of 16 page newsletter - Trinity Centre for High Performance Computing · 2017-12-12 · interconnect...
interconnectScientific Computing & Visualisation Newsletter Issue 3 December 09
Trinity College Dublin
Trinity Centre for
High Performance
Computing
interconnect
Dr. Geoff Bradley, Executive Director
(Acting), TCHPC, Trinity College Dublin.
Welcome to the third issue of Interconnect
- Trinity College Dublin's Computational
Science and Visualisation newsletter.
The Trinity Centre for High Performance
Computing (TCHPC) provides vital HPC
and Research IT services to the academic community in Trinity
and other third level institutes in Ireland through collaborative
projects and infrastructure grants. Services provided include
research support, advanced computing, storage, data
management, visualisation, research information systems,
hosting, software development, project management,
procurement and systems design and training. The Centre is
primarily funded by SFI, HEA and the EU.
To address the growing computational requirements of
researchers in materials science, bio engineering and
biochemistry, the centre has recently purchased an additional
11TF cluster, designed to support 1,024 core simulation runs.
Following a competitive tender process, Bull was chosen as the
supplier for this solution. This infrastructure investment was
funded by Science Foundation Ireland. This new cluster will be
integrated into the centres GPFS storage setup providing
researchers with common file access to 27TF of compute
resource across four clusters.
This issue of Interconnect:
This issue of interconect contains articles from researchers in
Physics, Medicine, Engineering and Chemistry. Dr. Patterson's
article covers his research of simulating high temperature
superconducting materials, Dr. Pinto describes some of the
activities of the neuropsychiatric genetics groups and the data
management and analysis required for genetic studies of
complex diseases. Prof. Rice and Mr. Clancy present details of
their work on accoustic shielding of aircraft noise by novel
aircraft configurations, while Dr. Ellis's article outlines his
research into the molecular dynamics of cell membranes and
the bridging of integral transmembrane protiens across them.
There are also a number of interesting infrastructure related
articles: Mr. Refojo provides details of latest advances in 3D
visualisation software under development in the Centre, Dr.
Garofalo (from CINECA, Italy) describes HPC-Europa2 - a pan
European Research Infrasturture of High Performance
Computing - which offers European Scientists the opportunity
to apply for access to first-class supercomputers and advanced
computational services, and Mr. Walsh provides an overview of
Grid Ireland. Finally, details of TCD's very successful M.Sc. in
High Performance Computing are provided on the back page. If
you've any comments or feedback on any of these articles, or if
you are interested in contacting the authors, please email us at
Contents
2
TCHPC – Building the nextgeneration of 3D Visualisation software
Molecular Dynamics of Biological Membranes:Investigating Biological SystemsMOLDY: An EU Transfer of Knowledge project.
Acoustic Shielding by Novel AircraftConfigurations
HPC-Europa2: Pan-European ResearchInfrastructure on High Performance Computingfor 21st century Science
Simulating High TemperatureSuperconducting Materials
Genetic Basis of Psychiatric Illness
Grid Ireland
M.Sc. in High Performance Computing
3
4
6
8
10
12
14
16
Introduction
interconnect
3
As computational power grows and datasets continue
increasing in both size and number, researchers are increasingly
turning to 3D visualisation as a tool to manage large datasets
and aid interpretation and analysis of results. TCHPC has long
appreciated the importance of 3D visualisation and has provided
visualisation services to the Irish research community since
2005.
These visualisation services fall into three areas: hardware
support, software support and software development. The first
area, hardware support, is currently managed by providing
researchers with access to TCHPC's visualisation facility. This
facility, which includes a fully immersive sterio 3D powerwall
with a tracking system, is currently being upgraded to provide
new state of the art compute and rendering resources.
The second area, software support, is generally handled on a
project-by-project basis as researchers can have very specific
datasets and visualisation requirements. In some cases these
requirements can be met by finding and deploying appropriate
third party visualisation software. However, it's often the case
that users' specific needs can not be met by using available out-
of-the-box software.
This highlights the importance of the
third area, software development. The
introduction of mature technologies,
such as GPGPU computing, and the
latest features in OpenGL, provide
functionality that is simply too useful or
too computationally powerful to ignore.
The need to keep up with these technological advances and
provide functionality to handle the increasingly complex
problems and datasets from the academic and business
community highlights the importance of having expertise to
develop 3D visualisation software.
Over the last five years, TCHPC has developed bespoke
visualisation software for a significant number of research
projects. Examples include, amongst others, stream line
visualisation (Foams) and isosurface building and rendering
(Astrophysics). This work has produced a significant amount of
code, bundled into a single piece of software, and provided staff
with a detailed knowledge of the most commonly occurring
visualisations problems and requested features from different
projects and research fields. This experience and expertise is
now being leveraged to build TVS – a next generation 3D
visualisation software – that takes advantage of technological
advances and recent trends in visualisation. This software will
make it possible for TCHPC staff to provide quality visualisation
support and best-of-breed software to enable scientists and
engineers to engage in ground breaking research.
TCHPC – Building the nextgeneration of 3D Visualisation softwareJose Refojo, Visualisation Specialist, TCHPC, Trinity College Dublin.
Electronic density visualisation of a lattice dataset.
interconnect
Biomembranes
Biological membranes play an essential role in every cell of all
living organisms. They perform the function of physical
container, permeability barrier and site of interaction for various
types of protein. These proteins fall into three classes; integral
transmembrane, which are fully embedded in the membrane
extending from the inside to the outside of the cell; integral
monotopic, which are partially embedded and located either
intracellular or extracellular; and peripheral, which are simply
associated with the membrane surface. Proteins classified into
one of these three groups have key similarities, despite their
diversity of structure and function. This is related to the amino
acids that are presented to the surrounding environment and
relates directly to the structure of the membrane.
The cell membrane, often referred to as the cell envelope
because of its double ply structure, are composed of two layers
of phospholipid molecules arranged hydrophobic tail to
hydrophobic tail (Figure 1).
Figure 1: The cell membrane. This figure has been taken fromWikimedia Commons(http://en.wikipedia.org/wiki/Image:Cell_membrane_detailed_diagram.svg)and is not under copyright.
This creates an impermeable barrier that prevents the passage
of polar substances, such as water, and ions both in and out of
the cell. The inner and outer surfaces of the membrane are
composed of the polar heads of the phospholipids which allow
the membrane to sit comfortably within a largely aqueous
environment. In order for integral transmembrane proteins to
Molecular Dynamics of Biological Membranes:
Investigating Biological Systems
MOLDY: An EU Transfer of Knowledge project.
Dr. Matthew Ellis, School of Chemistry, Trinity College Dublin
bridge the membrane whilst maintaining its integrity as a
permeability barrier, the surface is composed of a central
hydrophobic region. This is capped by two polar ends that are
exposed to the aqueous environment and interact with the
hydrophilic heads to hold it in place; integral monotopic proteins
have one polar region and one hydrophobic region that sits in
the membrane and acts as an anchor; whilst the peripheral
proteins only interact with the surface of the membrane, the
polar regions of other proteins and the aqueous environment,
and are therefore entirely polar.
GPCR's
In the pharmaceutical industry there is one family of proteins that
dominates the field, with the majority of drugs currently on the
market targeting them {1}. These are the G-protein coupled
receptors (GPCR), alternatively called 7 transmembrane
receptors (7TMR) due to the 7 highly conserved stretches of 22
– 28 hydrophobic residues that make up the central region and
characterise this family of proteins {2}. As integral
transmembrane proteins the GPCRs are in contact with both
sides of the cell membrane. Activation by an external stimulus
transmits a signal to the inside of the cell through a complex and
little understood activation mechanism {3}, which in turn
activates the heterotrimeric guanine nucleotide-binding
proteins (G proteins). These are peripheral proteins that
associate with GPCRs and act as secondary messenger in the
signal cascade.
The GPCRs compose a large group of proteins with ~30% of
the genome encoding them {4}. They detect a variety of stimuli,
such as hormones, growth factors, and even light, though there
are still a large proportion with no known endogenous ligand. Of
particular importance in the study of GPCRs are Rhodopsin,
which is activated by light, and β2-adrenoceptor (β2-AR), which
is modulated by noradrenaline and adrenaline. This is because
they are the only examples that have been solved by x-ray
crystallography {5}. Since the structure of β2-AR was solved
only 2 years ago it is still at the centre of a storm of research {3}.
4
interconnect
5
β2-AdrenoceptorThe initial research undergone during the MOLDY EU TOK
project concerned the development and validation of three new
all-atom membrane models {6}. These have been shown to
provide models that adhere to experimental observations in the
isobaric-isothermal (NPT) ensemble, which is a significant
improvement on previous models. The follow on from this initial
phase of the project involves the application of these three
models to the study of β2-AR. The structure of β2-AR
embedded in one of the membrane models is shown in Figure 2.
The primary aim was to show that the lipid models are not
adversely affected by the addition of the protein. Comparison of
metrics relating to the area per lipid and membrane width
showed that the addition of β2-AR has no effect on the
membrane models. This is shown in Figure 3 which depicts the
spatial distribution of lipid components and the depth of water
penetration.
Figure 3: Atomic distribution of lipids and solvent as a function ofmembrane width.
The importance of the development of these all-atom force
fields lies in the fact that interactions between the membrane
and protein occur at the atomic level. A prime example of this is
the interaction between the polar heads and aromatic belts {7}
that anchor the protein within the membrane. Figure 4 shows
snapshots taken every 100 ps aligned on the protein backbone.
Figure 4: Overlays of the conformations adopted by β2-AR for DPPC,POPC and PDPC systems.
These show that in all three systems β2-AR remains at a
constant orientation within the membrane. Further evidence
that each membrane model is of a consistently high quality is
found through comparison with β2-AR conformations in the
absence of a membrane. The RMSD from the crystal structure
of β2-AR is shown in Figure 5. It is clear that each of the
membrane models are exerting a consistent lateral pressure
which is key in maintaining the tertiary structure of β2-AR.
This successful application in the field of all-atom
membrane/protein simulations is an important step towards the
development of full cell simulations. With the current availability
of computer hardware it is only the development of theory such
as this that prevents such studies from being realized.
Acknowledgements
This work is funded by the EU Marie Curie Transfer of
Knowledge Programme. Resources were made available from
the HEA PRTLI IITAC Programme and provided by the Trinity
Centre for High Performance Computing.
References1. Terstappen, G. C. and Regianni, A. Trends Pharmacol. Sci. 2001, 22, 23-26.2. Hemley, C. et al. Current Drug Targets 2007, 8, 105-115.3. Dror, R. O. et al. Proc. Natl.
Acad. Sci. USA 2009, 106, 4689-4694; Huber, T. and Sakmar, T. P.Biophys. J. 2009, 47, (in press).4. Wallin, E. and von Heijne, G. Protein Sci. 1998, 7, 1029-1038.5. Palczewski, K. et al. Science 2000, 289, 739-745; Cherezov, V. et al.
Science 2007, 318, 1258-1265.6. Taylor, J. et al. Biochim. Biophys. Acta. 2008, 1788, 638-649.7. Domene, C. et al. J. Am. Chem. Soc. 2003, 125, 14966-14967.
Figure 2: β2-ARembedded in a PDPCmembrane and solvatedin 0.2 M NaCl. Lipidsand solvent has beencut back to show proteindetail.
DPPC POPC PDPC
Figure 5: RMSDfrom the crystalstructure of β2-ARfor MD simulationscontaining DPPC,POPC, PDPC andno lipid.
interconnect
Aircraft noise has long been a source of annoyance to residents
near air-ports. With the latest long term predictions forecasting
nearly 5% annual growth [1], there is mounting pressure on
aircraft manufacturers to produce significantly quieter aircraft.
Future low noise designs are expected to utilize the beneficial
noise shielding effect of the airframe. Several high shielding
configurations have been investigated under the Novel Aircraft
Concepts RE-search project (NACRE) and the Silent Aircraft
Initiative. The basic principle behind these designs is to position
the engine pods so that the airframe obstructs the direct path of
propagation between the engine noise sources and the target
area beneath the aircraft. Due to diffraction effects, a certain
amount of sound will always penetrate into the shadow zones,
however, with suitable design evaluation tools, a given
configuration can be optimized for maximum shielding effect.
At first glance, computing the shielding pattern of a novel
aeroengine configuration is a straightforward problem, once the
sources have been characterized, since the physics of sound
propagation and diffraction is well understood and was one of
the first problems to be tackled by numerical analysis.
In practice, the difference in scale between the characteristic
wavelength of sound and the characteristic length of the aircraft
structure, combined with the inhomogeneity of the acoustic
medium, makes for a challenging computational problem.
Even with large scale computing resources, current
computational aeroacoustics codes based on the Linearized
Euler Equations (LEE) are limited to low frequency calculations.
On the other hand, very fast methods exist for the scalar wave
equation and the Helmholtz equation in a homogeneous
medium, which are based on the Boundary Element Method
(BEM). The objective of this work is to assess a practical method
for shielding calculations that is similar to the fast methods with
respect to computational load but incorporates some of the
physical accuracy of the LEE codes.
Acoustic Shielding by NovelAircraft ConfigurationsMr. Cathal Clancy & Professor Henry Rice, School of Engineering, Trinity College Dublin
Contour plot of acoustic particle velocity computed by ACTRAN, showingvortex shedding at the wing trailing edge and Doppler shift of the soundwaves. (The XZ plane is the cutplane, with normal vector along the trailingedge, that bisects the wing).
Approximations for certain physical phenomenon that cannot be
modelled by the standard BEM have been added to our BEM
code. In particular, the refractive effect of the steady flow has
been modelled to first order in the Mach number. Other physical
processes such as the interaction of the sound field with the
steady vorticity and also the shedding of time varying vorticity
due to sound impinging on a sharp trailing edge have also been
investigated.
6
Representative high shieldingconfiguration with acoustic pressurecomputed by Boundary Element Method
interconnect
A test configuration consisting of a spherical compact noise
source shielded by a thin wing with a symmetric profile has been
used to assess the BEM approach. Reference CFD flow
solutions and LEE computations were computed using the
commercial packages FLUENT and ACTRAN. These
computations have been compared with an approximate
method based on the accelerated BEM for thin structures [2].
The results for the test configuration suggest that the highly
non-uniform flow modifies the near field sound field in a complex
way but has a relatively simple effect on the far field scattered
sound, as predicted by Taylor[3], which is adequately captured
by the approximate method. The time-harmonic vortex
shedding occurring at the trailing edge of the airfoil tends to
cause instability in a LEE simulation when the steady flow field
contains significant vorticity.
This feature is suppressed in the BEM approach, which results
in a more robust numerical approach for large scale
computations. The main benefit of the BEM approach is in
computational speed, with a reduction in CPU time of two or
more orders of magnitude when compared with LEE
computations.
Contour plot of acoustic particle velocity computed by BEM with flowcorrection, showing the vortex sheet extending from the trailing edge andDoppler shift of the sound waves. (The XZ plane is the cutplane, withnormal vector along the trailing edge, that bisects the wing).
References
[1] Airbus global market forecast: 2009-2028. Technical report, Airbus,Toulouse, France, 2009.
[2] C Clancy. Acoustic shielding in low mach number potential flow incor-porating a wake model using bem. In 15th AIAA/CEAS AeroacousticsConference, Miami, FL, USA, 2009.
[3] K Taylor. A transformation of the acoustic equation with implicationsfor wind-tunnel and low-speed flight tests. Proceedings of the RoyalSociety of London. A., 363(1713):271–281, 1978.
7
Contour plot of acoustic particle velocity computed by ACTRAN. (The XYplane is the cutplane that is parallel to the wing chord and the trailing edgeand also bisects the wing).
interconnect
8
High-performance computing (HPC) is essential to many
branches of science and technology, such as climate modelling
and aircraft design, but the partners in the HPC-Europa2 project
believe it has applications in many non-traditional fields too,
from life sciences to knowledge management and discovery. The
main function of this project is to give the European research
community access to first-class supercomputers and advanced
computational services in an integrated way. Anyone whose
work would benefit from HPC can apply for an all-expenses-paid
visit, lasting up to three months and including training and local
support. Through EU funding, HPC-Europa2 – and its
predecessors – has hosted hundreds of people in this way.
Number-crunching for newcomers
The rise of powerful computers has been good news for
scientists and engineers. Since the first supercomputers
appeared in the 1970s, researchers have been able to replace
some of their experiments with computer-based mathematical
simulations of the real world, often saving time and money in the
process. For many problems where real experiments remain
essential, powerful computers also help to increase the amount
of useful data that can be extracted from the results.
The original supercomputers were so-called ‘vector’ machines
with single processors – essentially beefed-up versions of
ordinary computers. By the 1990s, the supercomputing had
shifted to parallel computing, in which tasks are shared between
many processors, often similar to those found in desktop PCs. A
further development is the technique known as clustering, in
which a large number of separate processors or multi-
processors linked by fast interconnection networks cooperate to
create a powerful and reliable system scaling up to hundred
thousands of processors. Grid computing, which links computer
systems, supercomputers and instruments through the internet,
can also provide a collaborative environment to face complex
and computation-intensive problems.
Supercomputers are used for tasks such as weather forecasting,
climate research, modelling chemical compounds and biological
molecules, simulating the aerodynamics of aircraft and road
vehicles, particle physics, astronomy and code breaking.
In the current challenging scenario of the HPC eco-system in
Europe, the HPCEuropa2 project is improving Europe’s
competitiveness in R&D by making the best use of the most
advanced supercomputers, for instance MareNostrum in
Barcelona. HPC-Europa2 gives researchers across Europe
access to HPC within a high-quality computational environment,
including the necessary technical and scientific support and
specialist training, and is also helping to improve HPC facilities
generally.
A computing grand tour
The main objective of HPC-Europa2 is to continue providing a
high-quality service for transnational access to the advanced
HPC systems available in Europe. This activity has been
available on an ongoing basis as a highly rated and trusted
service for almost two decades. The project is organised around
its core activity – the transnational access HPC service
provision. Indeed, over its four-year lifetime, transnational access
will provide HPC services, specialist support, scientific tutoring
and opportunities for collaboration to more than 1 000 European
researchers. This very large community of users will be provided
with more than 22 million of CPU hours of computing time.
HPC-Europa2:
Pan-European Research Infrastructure on High
Performance Computing for 21st century ScienceDr. Francesca Garofalo, HPC-Europa2 Project Manager, CINECA
interconnect
9
Each visitor is guided by a host researcher, working locally in a
related field, who provides office space and a specialised
scientific tutoring. The supercomputing centres are CINECA
(Italy), EPCC (the UK), BSC (Spain), HLRS (Germany),
GENCICINES (France), SARA (the Netherlands) and CSC
(Finland).
HPC-Europa2’s transnational access activity allows any
researcher from an eligible country, whose work could benefit
from HPC, to visit one of a number of supercomputing centres
for up to three months, with all expenses paid. Over the course
of the project, hundreds of European researchers, from
postgraduates to senior professors, will benefit from this
opportunity.
In addition, a number of networking activities will be
implemented around the core business of the project; to interact
with the HPC ecosystem in Europe; to coordinate the
transnational access activities; and to coordinate the activities
related to user support, consultancy support and the diffusion
and dissemination of the HPC culture.
Three joint research activities are also being undertaken to
incorporate results of and contribute to the development of
emerging HPC programming models, to develop basic tools for
the scientific data service to improve the quality of information
extracted from the data and finally, to create a virtual cluster
environment which enables researchers to prepare and
familiarise themselves with the HPC environment in advance of
their visit, thus increasing the effectiveness and productivity of
transnational access visits.
Pan-European Research Infrastructure on HighPerformance Computing for 21st centuryScience
Project acronym: HPC-Europa2
Funding scheme (FP7): Integrating Activities (IA)
EU financial contribution: €9.5 million
EU project officer: Lorenza Saracco
Start date: 1 January 2009
Duration: 48 months
Project webpage: www.hpc-europa.eu
Coordinator: Francesca Garofalo, CINECA Consorzio
Interuniversitario, [email protected]
Partners:
� CINECA Consorzio Interuniversitario (IT)
� UEDIN-EPCC - Edinburgh Parallel Computing Centre (UK)
� BSC - Barcelona Supercomputing Centre (ES)
� USTUTT-HLRS - High Performance Computing Centre
Stuttgart (DE)
� GENCI-CINES - Grand Equipement National de Calcul
Intensif (FR)
� Centre Informatique National de l’Enseignement
Supérieur (FR)
� Commissariat à l'Énergie Atomique (FR)
� Centre national de la recherche scientifique (FR)
� SARA - Computing and Networking Services (NL)
� CSC - Tieteen tietotekniikan keskus Oy (FI)
� PSNC - Instytut Chemii Bioorganicznej PAN, Poznańskie
Centrum Superkomputerowo Sieciowe (PL)
� Unifob AS (NO)
� TCD - Trinity College Dublin (IE)
� ICCS-NTUA - Institute of Communication and Computer
Systems of the National Technical University of Athens (EL)
interconnect
Many of the most exotic phenomena in condensed matter
physics are seen in transition metal compounds. High
temperature superconductivity is perhaps the most intriguing of
these. These phenomena are sometimes modeled using
electron energy parameters derived from experiment or
‘guestimate’. The methods we apply using facilities at the Trinity
Centre for High Performance Computing (TCHPC) and the Irish
Centre for High End Computing (ICHEC) are known as ‘first
principles’ methods. Electron energies and other properties are
calculated with very few assumptions about the material under
study. Once the chemical identity of the atoms in the material
have been specified, properties as fundamental as atomic
positions in the crystal structure can be computed.
Superconductivity occurs when materials conduct electricity
without resistive losses of power. Many metals superconduct at
temperatures up to a few degrees above absolute zero. Over 20
years ago, physicists became very excited when
superconductivity was discovered in copper oxide compounds
called cuprates at temperatures well above liquid nitrogen
temperature (77K). ‘High temperature’ is obviously a relative
term when applied to superconductors. The record
temperature for cuprate superconductivity stands around 180K,
120K below room temperature. However liquid nitrogen is
inexpensive to manufacture and cuprates have some real world
applications. They are beginning to be used in power
distribution networks in the US, especially in cities where large
currents must be transported through conduits designed with
much lower power demands in mind than those of today.
Simulating High TemperatureSuperconducting MaterialsDr. Charles H. Patterson, School of Physics, Trinity College Dublin
Figure 1. (top panel) Schematic diagram showing electron magnetic
moments on Cu ions as arrows (small circles) between O ions (large
circles) in CuO2 atomic planes. O ions with a positive or negative
magnetic moment are shown as large coloured circles and are separated
by four lattice constants (ao). (middle panel) Magnetic moment density
on a CuO2 plane in a cuprate from a first principles calculation on
Ca2CuO2Cl2. (lower panel) Quantum mechanical wave function
amplitude mainly localised on Cu and O ions in the stripe. All three
panels are to the same scale.
10
interconnect
Superconductivity in metals at low temperature has been well
understood since the award of the Nobel prize to Bardeen,
Cooper and Schrieffer in 1972 [1]. The prize was awarded for
the explanation of how the interaction between electrons and
atomic vibrations in superconductors leads to electron pairing
as ‘Cooper pairs’ which can transport electric charge without
resistive losses. However the mechanism of high temperature
superconductivity in the cuprates remains a mystery, although it
has received much attention from distinguished physicists. The
author has recently completed work [2] on a feature found in
many cuprates under conditions where they are
superconducting, namely stripes. These are modulations in the
electron density on a length scale four times greater than the
fundamental cuprate unit cell and they appear as ‘stripes’ in
various experiments which are sensitive to charge density, such
as neutron scattering or scanning tunnelling microscopy [3].
Electrons possess both electric (charge) and magnetic (spin)
properties. In Figure 1 we show magnetic moment densities on
a CuO2 plane in the superconductor Ca1.87Na0.13CuO2Cl2,
schematically and from first principles calculations as well as
quantum mechanical wave functions amplitudes for ‘stripe
electrons’. The distribution of charge density has the 4ao
modulation observed in scanning tunnelling microscopy [3].
Quite recently physicists were surprised by the discovery of an
entirely new set of superconducting compounds known as
pnictides [4], whose crystal structure consists of Fe and As
layers separated by spacer layers containing, e.g. Ca ions. It is
of course interesting that both cuprates and pnictides contain
atomic layers (FeAs or CuO2) in which the supercurrent is
transported. In Figure 2 we show an isosurface of the
wavefunction for the electrons in CaFe2As2 closest to the Fermi
energy [5], i.e. those electrons with the highest energies. When
componds become superconducting, it is electrons at or very
close to the Fermi energy which form Cooper pairs and carry the
supercurrent. The isosurface in Figure 2 shows that the wave
function has Fe 3dxz and 3dyz character. Condensed matter
physicists are well acquainted with a construct known as the
Fermi surface, which is essentially the set of electron
wavelengths for all electrons which have the highest (Fermi)
energy in the material. Figure 3 shows the Fermi surface for
CaFe2As2 when some electrons have been removed by
chemical (hole) doping. It possesses two sheets in a shape
resembling an egg timer. Photoemission experiments [6] can
be used to measure the Fermi surface in materials and the
Fermi surface for pnictides such as doped CaFe2As2 is a two
sheet Fermi surface similar to that shown in Figure 3.
Figure 2. Isosurface of the wave function amplitude in CaFe2As2 forelectrons close to the Fermi surface.
Figure 3. Two views of the Fermi surface for hole doped CaFe2As2. Twosheets are visible. The inner sheet is coloured blue and yellow on theinner and outer surfaces of the sheet while the outer sheet is colouredgreen and purple.
Much more work needs to be done on these fascinating
compounds before the ways in which interactions between
electrons and atomic vibrations (or possibly other excitations of the
material) lead to formation of Cooper pairs and superconductivity.
Powerful computers in TCHPC and ICHEC are essential tools for
this type of work. They make it possible to study these effects
without having to resort to guestimates about the ways in which
electron interact. Continued increases in available computing
resources and developments in software for simulating these
materials will, no doubt, lead to much better understanding of their
properties at the atomic level in the near future.
[1] http://nobelprize.org/nobel_prizes/physics/laureates/1972/[2] C. H. Patterson, Phys. Rev. B 77, 094523 (2008).[3] Y. Kohsaka et al., Science 315, 1380 (2007).[4] C. Xu and S. Sachdev, Nature Physics 4, 898 (2008).[5] C. H. Patterson (unpublished).[6] http://physics.aps.org/articles/v1/21
11
interconnect
Genetic Basis of Psychiatric IllnessDr. Carlos Pinto, Neuropsychiatric Genetics Group, Institute of Molecular Medicine,
Trinity College Dublin
Genetic Association Studies
We are currently involved in the following collaborative genome
wide association studies (GWAS) of schizophrenia, ADHD and
autism:
The Group is a member of the International Schizophrenia
Consortium (ISC) that is currently undertaking a GWAS of
schizophrenia at the Broad Institute, Boston, USA. The study has
genotyped a European sample of 3,600 cases and 4,200 controls
at a million locations across the genome. We are also members of
the Wellcome Trust Case Control Consortium Phase 2
Schizophrenia GWAS.
We are founding members of the International Multicentre ADHD
Genetics Study which was selected as one of six studies to
participate in the seminal NIH GAIN GWAS. The genotype data for
this study (600,000 markers in 958 families) is now available. We
have installed the data on our local systems and completed the
Stage I analysis locally.
We are members of the Autism Genome Project (AGP), a multi-
centre study representing the largest international collaboration of
autism researchers, with data at a million locations in 1,000 families.
Genotyping is being conducted at several sites with the Irish site (a
TCD-UCD collaboration) undertaking the greatest proportion of this
work.
12
Many psychiatric conditions are known to have a genetic
component. In most cases, the underlying genetic structure is
complex. The systematic investigation of these conditions has now
become feasible due to advances in genotyping technology, which
have revolutionised the study of the genetic basis of human
disease.
These new technologies generate data on a scale several orders of
magnitude greater than was possible even a decade ago; this, in
turn, requires the development of new approaches to data
management and analysis. Modern genetic studies of complex
disease typically involve major international collaborations, with
many thousands of individuals being genotyped at hundreds of
thousands of locations across the genome.
A recent development has been the advent of Next Generation
Sequencing (NGS) technologies. Traditional sequencing methods
represented a considerable bottleneck in the research of many
biological phenomena, both in terms of cost and the lack of true
high-throughput production pipelines. In contrast the new
technologies utilised by NGS platforms provide an individual
laboratory with the sequencing capabilities traditionally associated
with large scale genome centres.
The Neuropsychiatric Genetics Research Group is involved in all
aspects of these studies, including the collection of clinical data,
genotyping and sequencing of DNA, and analysis and
interpretation of the results.
In order to support this research we have established a dedicated
data management and analysis system. This is based around five
high end Linux based servers and 50 Tb of storage capacity and is
hosted at the Trinity Centre for High Performance Computing. We
also have access to the TCHPC computing clusters.
interconnect
Cognitive Genetics
The Cognitive Genetics Laboratory focuses on understanding how
illness risk is increased by specific genes, through focusing on
specific aspects of brain function. Our work draws on
neuropsychological, electrophysiological, and neuro-imaging
techniques for investigating the role of gene function at the level of
individual brain systems. Measures of neuropsychological ability,
include general cognitive ability (IQ), memory, and attention.
The use of high density EEG to study variance in sensory
information processing, both at early and late processing stages
involves a non-invasive measurement of electrical impulses picked
up by scalp electrodes. Neuroimaging approaches involve the use
of MRI for a wide range of purposes, including measurement of grey
and white matter density, white matter integrity (DTI), and functional
MRI (fMRI). Collectively, these provide millimetre accuracy in
investigating the influence of individual genes on brain structure
and function.
Employing these approaches to understanding the dysbindin gene,
(a candidate gene for schizophrenia) we have shown that risk
variants of the gene, are associated with poorer cognitive function
and sensory level processing, and concomitant reduction in brain
volumes in the relevant brain areas. In this way these studies
contribute to understanding the functional role of these genes and
the likely mechanism by which illness risk is being conferred.
Next Generation Sequencing
The Trinity Genome Sequencing Laboratory operates an Illumina
Genome Analyzer II (GAII), a second generation DNA sequencing
platform. This high throughput system is currently being used to
carry out a broad range of genetic analysis including investigation of
genetic variation within a species by whole genome resequencing,
profiling of gene expression differences between diseased and
non-diseased individuals within a population as well as
characterisation of the DNA regulatory mechanisms within a cell,
based on protein-DNA interactions.
The Illumina Genome Analyzer can now generate between 1-2
terabytes of data every five days. The analysis of such vast amounts
of data is computationally intensive whilst data storage also
represents a considerable problem. A collaboration between the
sequencing laboratory and TCHPC has facilitated the generation of
a suitable analysis pipeline capable of both efficient analysis of, and
access to the data generated by the Genome Analyzer as well as
robust long term storage of the DNA sequence data.
Bioinformatics and Statistical Genetics
In addition to providing bioinformatics, statistical and computational
support to the Group, the bioinformatics team conducts research
on novel methods of data analysis appropriate to the size and
complexity of the datasets that are now becoming available. These
include the application of Bayesian statistical techniques and
machine learning approaches to the analysis of complex genetic
traits and require significant computational resources.
Website
http://www.medicine.tcd.ie/neuropsychiatric-genetics/
Research Funding
SFI, HRB, Wellcome Trust, NIMH
Sample Publications
O’Donovan et al.Identification of loci associated with schizophrenia by genome-wideassociation and follow-up. (2008) Nature Genetics 40: 1053-5
Ferreira et al.Collaborative genome-wide association analysis supports a role for ANK3and CACNA1C in bipolar disorder. (2008) Nature Genetics 40: 1056-8.
GNeale BM, Lasky-Su J, Anney R, Franke B, Zhou K, Maller JB, Vasquez AA,Asherson P, Chen W, Banaschewski T, Buitelaar J, Ebstein R, Gill M, MirandaA, Oades RD, Roeyers H, Rothenberger A, Sergeant J, Steinhausen HC,Sonuga-Barke E, Mulas F, Taylor E, Laird N, Lange C, Daly M, Faraone SVGenome-wide association scan of attention deficit hyperactivity disorder.
Am J Med Genet B Neuropsychiatr Genet. 2008 Dec 5;147B(8):1337-44.
13
interconnect
14
Grid-Ireland was established in 1999 to develop and coordinate
the provision of a national grid service for the academic research
community in Ireland. It is recognised internationally as the Irish
co-ordinating body for grid activities in major EU projects, such
as the EGEE and EGI. It is the body that issues grid certificates
accepted by most major grid infrastructures.
The Grid Operations Centre (OpsCentre), based in Trinity College
Dublin, manages the Regional Operational Centre (ROC) for
Ireland. It is funded under the national research initiatives HEA
PRTLI cycle 4 and by the EU FP7.
What is a Grid?
Grid Computing is a generic term used to describe a distributed
ensemble of services and resources which enable communities
of users to form Virtual Organisations (VOs) in a co-ordinated,
structured, controlled, secure and trusted manner. The resource
providers control who may access their resources, be that
access to compute cycles or data storage. The goal is to
seamlessly enable Grid users access to a diverse range of
distributed resources in the most efficient and suitable manner
possible.
The EGEE Grid
TCD/Grid-Ireland is a partner in the EU FP7 Enabling Grids for
eScience (EGEE) project . Currently, this grid infrastructure is
composed of some 310 sites from 57 countries, providing over
80000 CPU cores and 81 petabytes of data storage.
Application Areas
Grid-Ireland currently supports users, VOs and applications in
the following scientific domains:
� Bioinformatics
� High Energy Physics, including the Large Hadron Collider
project at CERN
� Geophysics and Earth Sciences
� Astronomy and Astrophysics
� Computational Chemistry
� Network Simulation
� Mathematical Research
� Grid Middleware Development
� Marine Sciences
Grid applications are typically loosely coupled, where the
workload can be easily broke-down into smaller sets of
individual tasks that have little or no interaction with one another.
In this way, one may distribute a large job over a set of
distributed resources in an efficient manner. In addition, Grid-
Ireland researchers based in TCD have helped expand the
support for MPI-enabled on the EGEE Grid. A standard
mechanism for converting MPI applications into Grid-aware MPI
applications is well documented and supported.
User and Application Support
The OpsCentre is dedicated to helping users design and grid-
enable their applications. Applications Support scientists will
advise on the best way for users to get their applications up and
running as quickly as possible. A helpdesk provides a standard
means of tracking all user requests and support issues.
Grid IrelandMr John Walsh & Dr Brian Coghlan, School of Computer Science and Statistics,
Trinity College Dublin
interconnect
15
Grid-Ireland Infrastructure
Grid-Ireland has a point of presence at 17 HEA institutions in
Ireland.
The TCD OpsCentre centrally operates a national Virtual
Organisation Management Service (VOMS), a grid job
submission service, information services, and data storage and
catalogue services. Our outreach and dissemination effort
provides Migrating Desktop and P-Grade portals, as well as a
selfpaced eLearning system, user support and user application
development, and user helpdesk. The national grid Certificate
Authority (CA) is supported by two regional authorities, one
each in UCC and NUIG. The OpsCentre is currently
investigating how to offer TCD users grid credentials by
integrating this with institutional federated management
services, thus making the process of accessing national grid
resources much easier.
The OpsCentre also leads the effort for middleware platform
portability in the EGEEIII project, and is also currently
participating in a national grid-enabled data service.
Integrated e-Infrastructure
In September 2007, the e-INIS initiative, involving the Irish
academic institutions and major network and computational
infrastructure providers was begun. Its mission is to coordinate
and enhance activities to create a sustainable national
einfrastructure offering: a core High Performance Computing
(HPC) facility to all Irish third-level communities; a single point
access to integrated computational, data and network
resources; provision of these resources at a scale that is
internationally competitive; access to high level expert user
support and training; and a recognised Irish nexus for
international e-infrastructure initiatives.
The key expected outcome is an integrated national e-
infrastructure which includes: availability to the entire
community; capacity and capability computing; a pilot data
service; secure network and grid services; specialist high-level
support; and a management and business plan for long-term
support. e-INIS runs until December 2010.
Towards the European Grid Infrastructure
Grid-Ireland is the National Grid Initiative (NGI). It is a founder
member of the European Grid Infrastructure (EGI). With the
participation of 37 EU Economic Area states, the EGI offers will
oversee the unification of several large grid infrastructures in
Europe, allowing several diverse grid middleware to interoperate
with each other.
Contact Details
www.grid.ie,
Irish NGI representative:
Dr Brian Coghlan,
School of Computer Science and Statistics,
Trinity College Dublin, Dublin 2, Ireland.
Tel: +353-1-8961766
e-mail: [email protected]
Irish NGI Alternate:
John Walsh
School of Computer Science and Statistics,
Trinity College Dublin, Dublin 2, Ireland
Tel: +353-1-8961797
e-mail: [email protected]
interconnect
16
The TCD School of Mathematics supports advanced, practical
training in High-Performance Computing through its continued
provision of a taught M.Sc. programme. The degree was
established in 1997 and has a long record of delivering training
in the use of systems with many compute cores in large scale
numerical investigations.
High Performance Computing techniques continue to find
applications in more and more fields. Numerical simulations
and analyses of large datasets are widely used in research in
physics, mathematics, chemistry and biotechnology as well as
engineering and finance. The use of detailed and increasingly
realistic mathematical models of processes and systems that
are difficult or expensive to study in reality requires similarly
large-scale computing resources. Often a reliable description
needs both a lot of numerical computing and data storage and
manipulation and the hardware and software components must
be coupled and used efficiently.
The course aims to train students in applications of advanced
numerical simulation in industry, finance and research.
Students learn both programming skills and the mathematical
foundations that enable them to carry out large numerical and
data intensive studies of complex systems found in scientific
and technical domains. The M.Sc. programme is built around
core topics that ensure students understand how to solve large
numerical problems on modern HPC architectures. These
courses cover aspects of computer architecture, parallel
programming, software profiling and optimisation, classical
simulation algorithms and stochastic modelling. As students
gain experience in these areas, they simultaneously develop
more specialist skills in an application domain, such as a
scientific discipline or financial modelling.
M.Sc. in HighPerformance ComputingDr. Mike Peardon, School of Mathematics, Trinity College Dublin
A strong hands-on element is central to the programme. In
most courses, assessment is through a set of programming or
numerical exercises that are designed to illustrate material
covered in lectures. To solve these problems, students have
access to a dedicated teaching laboratory and can run the
programs they develop on the facilities provided by the Trinity
Centre for High Performance Computing (TCHPC) which
include large parallel computing clusters which use a variety of
processor and network architectures. To complete their
training, students are required to report on a substantial project,
which involves them solving a large technical problem in a
research or technology domain of their choice. A number of
these projects have been carried out in partnership with HPC
users or providers, such as IBM Dublin.
In recent years, a partnership with a similar programme run by
the Faculty of Mathematics and Natural Sciences at Wuppertal
University in Germany has been established. A joint HPC
seminar programme, held via a video conferencing system has
been running successfully for two years. A longer term aim is to
expand this partnership across the EU.
A trapped quark. Monte Carlo simulations of the strong interactionreproduce the phenomena of confinement; the quarks that make up theproton and neutron are never seen alone. These calculations use some ofthe largest parallel supercomputers in the world.
Contact details:
http://www.maths.tcd.ie/hpcmsc