|
|
Research Fellow Job ref 210554
School of Computing
Closing Date: 5-09-2005
Further details:
Self-Adapting Software for Grid-Based Numerical Simulation
The post is available for a fixed term of three
years to work on an EPSRC grant entitled "Self-Adapting Software for
Grid-Based Numerical Simulation" recently awarded to Professor Peter
Jimack as part of the EPSRC's Computer Science for e-Science call. You
will be working in the Scientific Computing group within the School of
Computing at Leeds, a group that has a record of
internationally-leading research in parallel numerical algorithms. You
would also be collaborating closely with the Informatics Architectures
group in the School of Computing at Leeds who lead our research in Grid
Computing.
Research IA (£19,460 – 29,128 p.a.) depending upon experience
The University is introducing a new reward
framework which will facilitate the recruitment, retention and
motivation of world class staff.
Information about the School of Computing can be obtained from http://www.comp.leeds.ac.uk
Informal enquiries should be made to Prof. Peter Jimack, pkj@comp.leeds.ac.uk http://www.comp.leeds.ac.uk/pkj tel +44 (0) 113 343 5464.
Application packs from Irene Rudling tel 0113 343 35480 email i.rudling@leeds.ac.uk
Job ref 210554 Closing date 5 September 2005
Background
This project is concerned with the effective
utilisation of Grid computing within one of the core components of
e-Science: that of numerical simulation. In particular it focuses on
developing and understanding fundamental techniques that will allow
numerical applications to automatically adapt to their context. In
order to ensure that the research remains focused we propose to
consider one particular, but extremely important, class of numerical
techniques based upon finite difference (FD) and finite element (FE)
solution of partial differential equations (PDEs). Furthermore, we will
specifically target the use of the most modern of sparse algebraic
solution algorithms based upon multigrid and related methods.
Computational Grids should enable the effective use
of geographically distributed resources by distributed teams of
researchers in a transparent and seamless manner. Their potential has
already been successfully demonstrated across a very wide range of
scientific applications involving distributed data analysis, remote
visualisation and large-scale computation. It is the latter aspect of
Grid computing and e-Science that is to be addressed by this project.
In particular it will focus on the development of new techniques and
algorithms to allow such e-Science applications to access the wide
variety of heterogeneous compute resources potentially available on a
Grid in a manner that combines maximising the efficiency of their
utilisation with maintaining full transparency for the scientific
users. There are many potential techniques and parameter choices
associated with the parallel implementation of numerical software for
PDEs and this project will focus on better understanding how such
software can automatically and intelligently adapt to make the best
possible use of available resources.
Job summary
This proposal is funded as part of the EPSRC's
second call for proposals for Research in the Computer Science
Challenges to Emerge from e-Science. The key scientific issues to be
addressed by this project relate to the development of techniques and
algorithms to produce automatically adapting software for parallel and
distributed computing on computational Grids. Amongst others, this will
require the development of: dynamic load-balancing procedures that are
able to respond to observed behaviour on a given computational node;
different partitioning and parallelisation techniques that are best
suited to different architectural characteristics; reliable
computational models, whose parameter values may be easily established
at run time, in order to predict the most efficient execution format
for the hardware encountered, and; novel algorithms for dealing
efficiently with the high latency issues arising when simulations are
undertaken across more than one Grid resource. The specific project
objectives are as follows.
· To understand the computation, communication and
memory access patterns of a range of common parallel FD and FE software
when executed across a variety of architectures on a computational Grid.
· To develop techniques and algorithms that will
allow such software to automatically adapt to a given architecture so
as to improve the computational performance on that architecture,
whether it be homogeneous or heterogeneous.
· To allow such software to execute on a single
problem across more than one Grid node in a manner that automatically
adapts the algorithm and its associated parameters to the available
computational and network resources in order to optimise performance.
· To ensure that execution models can be created in
order to allow reliable predictions to be made as to the performance of
these FD and FE algorithms, so as to allow the optimal scheduling of
multiple runs across different data sets, and to automatically adapt
the schedules in a dynamic manner during execution.
· To develop job description metrics to allow the
best resource allocation to be achieved for a given job and to define a
scheme to record details of the provided resources, and the algorithmic
and parameter choices made, in order to provide records of provenance
for data produced.
· To demonstrate the performance of the novel
algorithms developed in this project on realistic computational Grids
such as the White Rose Grid (WRG).
You will be responsible to the project investigator,
and will be working to achieve the objectives set out above. You will
be expected to work closely with colleagues in the Scientific Computing
and Informatics Architectures groups.
Person Specification
It is essential that you possess a PhD in a relevant
subject, or have the equivalent experience. You should also possess the
following skills and experience.
Essential:
· Expertise in some aspects of scientific computing and numerical algorithms.
· Excellent programming skills.
· The ability to write and present the results of research.
· Intellectual maturity and the willingness and ability to interact with experts across traditional domain boundaries.
· An ability to work independently but with appropriate guidance and working towards agreed goals.
Desirable:
· Experience with parallel numerical algorithms and programming.
· Experience with grid applications programming.
· Good knowledge of some or all of the numerical
methods that underpin the project (finite difference schemes, finite
element methods, multigrid methods and domain decomposition methods).
How to apply:
Applications should include the following:-
· A completed application form
· Equal Opportunities Monitoring Form . Please return the Form in a separate envelope marked 'EOs Monitoring'.
How to Apply
Applications should include the following:-
· A completed application form
· A Curriculum Vitae/information requested on page 2 of the form
· Equal Opportunities Monitoring Form (Enclosed). Please return the
Form in a separate envelope (enclosed) marked 'EOs Monitoring'.
Replies will be treated in complete confidence.
Completed applications should be returned to Professor Peter Jimack,
School of Computing, University of Leeds, Leeds, LS2 9JT quoting job
ref 210554 not later than 5 September 2005
If you are selected for interview you can expect to hear from the
University not later than 4 weeks after the closing date. If you are
not selected for interview the University will not contact you again.
A Criminal Records Disclosure is not required for this position.
Disabled Applicants
The post is located School of Computing. Disabled applicants wishing to
review access to the building are invited to contact the department
direct. Additional information may be sought from the Team Co-ordinator
in Disability Services, email disability@leeds.ac.uk or tel 0113 343
3927
Disabled applicants are not obliged to inform employers of their
disability but will still be covered by the Disability Discrimination
Act once their disability becomes known.
Data Protection
The information you provide in your application will be used to
consider your suitability for the post for which you have applied. If
your application is not successful the information will be disposed of
confidentially within 8 months. If your application is successful and
you are appointed, your information and future data will be processed
in accordance with the University's Data Protection Code of Practice. A
copy of this code can be obtained from either the University's Human
Resources Department or by visiting http://www.leeds.ac.uk/hr/policy/index.htm
Health and Safety Responsibilities
You are required to adhere and comply to the provisions of the Health
and Safety at Work Act, related Regulations and in accordance to the
University's Policy on Health and Safety which can be accessed via http://www.leeds.ac.uk/safety/usp/uspindex.htm
In addition you are also required to cooperate with regard to the
implementation of Health and Safety arrangements and should not
interfere with or misuse anything provided in the interest of Health,
Safety and Welfare at Work.
|
|
|
|