Information Systems
Society has become accustomed to the ever-increasing capabilities of information technology systems, which have become ubiquitous in managing and enhancing modern life. The Information Systems Research Division is at the forefront of exploring how information systems will evolve in the future to meet these expectations. Research is targeted at exploiting the increasing power of computers ? through development of new applications that push the boundaries of science-and through research into how networks of computers will lead to new distributed systems that dynamically adapt to the requirements placed on them.
Research Themes
Distributed Services
FLE carries out research into all aspects of distributed computing, focusing on the services context. Current efforts focus on the application of Service Oriented Architecture to Web Services in support of self-managed resources. The approach to Web Services development relies on the “stateful” model of Web Services, rendered using the Web Services Resource Framework specification published by OASIS. Future plans aim at applying the self-management approach to the design, development and deployment of services using novel techniques to ensure that these new services are “self-management ready” from the start.
FLE has an active and influential role in several standards development organisations: the Open Grid Forum, OASIS, W3C, and DMTF. Contributions are made at both the technical and managerial levels and FLE has a portfolio of a couple of dozen standards to which it has made significant contributions.
The Distributed Services group in FLE is one of the pioneers of Grid Computing, the results of which have now become mainstream distributed computing practice. Starting in 1997 members of the group led the design and development of the Unicore Grid infrastructure. Unicore is an open source project now involving a number of institutions and is used operationally in many deployments.
Current and Recent Projects
Scalable Applications
Within the next few years the world’s fastest computers will be able to perform a million billion arithmetic operations every second, breaking the petaflop/s barrier. This next generation of high-performance computers will consist of many thousands of multi-core processors, and this poses considerable challenges for the development of efficient applications software. FLE is collaborating with Fujitsu Limited’s Next-Generation Technical Computing Unit and with European partners such as the University of Oxford to develop software to exploit petascale-class computers.
Heart Modelling
Many drugs fail to reach the market because of side effects on the heart. Detection of such side effects late in the development process can lead to a waste of over 10 years of R&D effort and over one billion euros of R&D expenditure to the pharmaceutical company concerned. Computer simulation through in silico drug trials offers the chance to streamline the drug development process, with drug candidates approved earlier and at a much lower cost and with unsuitable compounds being detected much earlier. Computer simulation will also lead to reduced risks from clinical trials, will minimise animal testing and offers a step towards personalised healthcare. However, simulation of the electrophysiology of the whole heart is not currently feasible because of its enormous computational requirements.
FLE is working within a European Union funded project called preDiCT, one aim of which is to achieve whole-heart modelling in faster than real time. In combination with partners at the University of Oxford and at CRS4 Bioinformatica in Sardinia, we are developing new algorithms to exploit adaptive meshing combined with efficient parallelisation for petascale computers to improve the performance of current software by many orders of magnitude.
Biomolecular Simulation
Recent work at FLE has led to the development of a new simulation methodology based on a hybrid of Monte Carlo and molecular dynamics. This method, called Generalized Shadow Hybrid Monte Carlo (GSHMC), offers a rigorous, flexible and efficient approach to conformational sampling in biomolecular simulations. In GSHMC, a modified Metropolis criterion and a flexible momentum update significantly improve the acceptance rate compared with standard hybrid Monte Carlo simulation and at the same time allow dynamical information to be retained, making GSHMC a powerful tool for simulation of large molecules.
A recent collaboration with the group of Professor Mark Sansom in the Department of Biochemistry at the University of Oxford has provided a demonstration of the efficiency of the GSHMC approach. Simulations for a coarse-grained representation of a small peptide toxin interacting with a phospholipid bilayer show that GSHMC allows for a quicker localization of the toxin to the head-group/water interface of the bilayer by an order of magnitude relative to molecular dynamics.
GSHMC is well suited to high-performance computing. Parallelism in the momentum update step and in the molecular dynamics trajectories can be combined with any inherent parallelism in the application of interest and with other enhanced sampling techniques to achieve high levels of efficiency.
