Hybrid Computational Infrastructure Supporting eResearch

Important developments -the data deluge, Cloud computing, multicore architectures and growing importance of lightweight clients (tablets and smartphones) -are changing the Cyberinfrastructure (eInfrastructure) supporting eResearch. Many of the detailed features of Grids seem unlikely to survive with clouds replacing many aspects of compute grids. On the other hand, supercomputers and clusters supporting traditional parallel computing will in near term at least remain unchanged. Those aspects of Grids supporting data will still be needed as data is intrinsically distributed as it is gathered by a multitude of sensors and instruments. However the utility computing aspects of Grids including high throughput computing seem likely to move to clouds which offer excellent support for loosely coupled jobs not requiring the low latency and localization of (MPI-based) parallel jobs. As described below new computing paradigms typified by MapReduce appear attractive as they offer ease of use with little performance loss. As noted below, the importance of clouds and the new software ideas is not just their technical merit but also that they are supported by commercial software and so it appears more likely that cloud-based eInfrastructure will be easier to sustain than Grids where lack of commercial interest implies reliance on academic software.