e-Health Connectivity to UKLight
Partners: Oxford University, National e-Science Centre, UCL and CCLRC
This project addresses several important issues affecting deployment of Grid-based e-Health applications over optical networks including reliability, security, manageability, data delivery and quality of service. These issues will first be reviewed and requirements and constraints documented before investigating the practical benefits to the Integrative Biology project from using UKLight.
The computational approach to modelling the behaviour of complex biological systems, such as the heart, being pursued in the Integrative Biology project involves generating and manipulating large datasets of 10-100 terabytes in simulations using state of the art HPC facilities. To support this, the project is developing a toolkit of services including computational steering, collaborative visualisation, performance control, data curation and data mining.
The volume of data being generated, and the geographical distribution of the partners across the UK, Europe, New Zealand and the USA, mean the project will benefit greatly from access to the improved network capability offered by UKLight. Use cases being studied in the project include data transfer, visualisation and coupled simulation. Two particular end-user applications are:
- Performance control - investigating the use of heterogeneous HPC resources within a single simulation. In modelling heart mechanics, the mechanical pumping action of the heart is modelled using a relatively coarse, high-order finite element mesh on which to solve the non-linear finite deformation elasticity equations. These equations are coupled to the electrical activity generating the muscular contractions via a system of up to 50 ordinary differential equations (ODEs) which model the ion flows through the cell membrane. These ODE systems are solved on a much finer spatial grid with up to 30 million nodes but are only loosely coupled spatially. A potentially attractive approach, therefore, is to solve the ODE systems on an MPI HPC architecture whilst simultaneously solving the mechanical problem on an OpenMP architecture. Making this approach feasible will require high bandwidth communication and the ability to co-schedule the simulations on the different HPC resources.
- Visualisation - at present researchers store all their data on local disk to achieve low latency and reliable access. In future, higher resolution simulations will produce more fine-grained data in volumes will cannot be stored on a local machine. The project is therefore investigating new caching techniques to support real-time visualisation which are both reliable and have low latency.