Earth Sciences & Env.
Engineering & Tech.
Information & Comm. Tech.
Life Sciences & Biotech
Use this search facility to find out more about the profile of our HPC-Europa2 visitors, the type of work they have been doing, and their project achievements.
1. Development of Essential Molecular Dynamics Algorithms, integrating theoretical frameworks developed semi-independently in Nottingham and Barcelona.
2. Code development, within fortran9x/MPI framework
3. Evaluation, comparing results for DNA microcircles with previous data.
4. Analysis of code efficiency and scaling behaviour.
5. Initial proof-of-concept studies on much larger DNA circles, up to 1000 base pairs in size.
1. Working closely with the Orozco group, the basic Essential Molecular Dynamics algorithms were constructed. Much time was spent working out how to implement a Langevin Dynamics framework, and how to correctly integrate internal and external (intermolecular) forces in this novel approach.
2. The Essential Molecular Dynamics software was written "from scratch" in fortran95/MPI. The software has a completely different structure to a standard MD code, as it uses an agent-based approach. Each agent (processor) is responsible for calculating the dynamics of one molecule in an almost completely autonomous way, with non-bonded interactions being handled by direct inter-agent communications. There are no MPI collectives. As a result of this structure the code scales almost perfectly linearly with system size (though more work is needed to evaluate performance fully).
3. Simulations on alternating-GC and alternating-AT 90-base pair DNA microcircles were performed, at various degrees of positive and negative superhelical stress. Each simulation was performed on 91 processors (effectively one for each base pair, plus one to handle synchronisation and I/O). The circles showed very similar behaviour to what has been observed previously in conventional atomistic simulations, however the present simulations are much faster. Circles with no or just a small amount of superhelical stress remain open, however highly over- and underwound circles show supercoiling. This part of the project took longer than expected, as many test runs were required to establish the best way to handle non-bonded interactions and ensure the simulations remained stable.
4+5. Unfortunately time and computer time have run out before code efficiency and scaling behaviour can be analysed in detail. In addition, it has not yet been possible to test the method on very large systems where more than one thousand processors might be required.
The main objective of this project is to develop and test a new middleware tool for a parallel compiling system. Our current system aims to automatically introduce in the code efficient data partitions and scheduling techniques, guided by platform information. The new middleware will integrate in the system different synchronization and communication patterns found in parallel skeletons, that will help to automatically adapt the computations to the actual architecture and topology of the target machine.
This work is part of the Trasgo project. Trasgo is a source-to-source compiler system that translates simple high-level specifications of parallel algorithms to lower-level native programs, with data partition and communication details generated automatically. Hitmap is the run-time library used by the back-ends of Trasgo, for hierarchical tiling and mapping of arrays or sparse data structures.
Parallel skeletons present an interesting approach for parallel programming. Many parallel-applications present specific synchronization structures which are common for many of them. The patterns or skeletons can be instantiated with different parameters and functionality, hiding to the programmer the details of the internal processes management. The synchronization requirements are particular for each skeleton, and thus, it is easier to adapt its mapping techniques to deal with new architectures or complex topologies. On the other hand, they provide a limited set of programming structures with very different approaches and parameters. It is the responsibility of the programmer to understand each of them, and to combine them properly to build a complex parallel application.
Our proposal is to generate an extension of the Hitmap library, introducing a new abstraction level that bridges between synchronization and communication structures detected at the application level, and the Hitmap primitives for communication. This bridging model will be based on parallel skeletons. Thus, the analyzer of the higher level code will be able to easily derive efficient implementations on top of Hitmap with minimal transformation effort. The resulting codes will rely on the new skeleton driven middleware and the Hitmap mapping system to adapt the data partition, mapping, and synchronization to the current architecture and topology of the actual machine.
First, we should evaluate existent skeleton libraries for their compatibility with the current Hitmap system. Second, the load-balance and mapping techniques should be introduced as plug-ins in the Hitmap library, and their synchronization and communication structures should be expressed in terms of Hitmap functionalities. Then, skeletons could be integrated as modules on top of Hitmap, providing a proper middleware for the Trasgo code generating system.
We have developed a new middleware tool for our parallel compiling system. It integrates different synchronization and communication patterns found in parallel skeletons. It will help our system to automatically adapt the computations to the actual architecture and topology of the target machine.
Development of parallel software is a quite complicated task. Parallel skeletons take into account that a great number of parallel algorithms have a well-know patterns of computation and iteration. Skeletons are a high-level parallel programming model that specify the overall structure of common computation patterns, hiding the complex details of parallel applications.
Our current system, Trasgo, is a source-to-source compiler that translates simple high-level specifications of parallel algorithms to lower-level native programs, with data partition and communication details generated automatically. Hitmap is the run-time library used by the back-ends of Trasgo, it is a library for hierarchical tiling and mapping of dense arrays.
The achieved results obtained during this project are:
The objectives of this research program are the following:• Production by the application of integrated computational models with multiscale approach based on data from the literature, thus providing a basis for future research. The verification of the developed models will be achieved through the comparison with literature data or with a synergistic collaboration with the experimental sector. The models will also be applied to the selected pure materials • Using simulations on various scales for the detailed analysis of the distribution of voids and the morphology of high free volume polymers (HYFLON, TEFLON, PIM-PIs). • Identify at least two different nanofillers by analyzing the structure at the molecular level and morphology of the materials under study: the chosen particles are indicated to be suitable for the preparation of mixed matrix membranes (MMM) for specific pairs of gas separations.• Calculation of theoretical values of solubility, permeability and diffusion of gaseous species to be separated, together with detailed analysis of the interactions on the nano- and micro-scales of permeant with nanoparticles and polymer matrix.
takin into account that the period of my visit has been reduced to only one month the achievements are reduced compared to the expected results. Meanwhile I was able to achieve the following objectives:
1) Investigation of the morphology of complex copolymer systems (Hyflon ion for fuell cell application) with a combined Molecular Dynamics/ Mesoscale modelling approach by using the new software Culgi 7.0. developed in the host research group and interfcaed with the Huygens supercomputer of SARA.
2) Validation of the DDFT simulated models by comparison with experimental data.