We consider Grid computing to be mostly a solved problem now. The Distributed Resource Managers such as DataSynapse Grid Server and Platform Symphony provide sophisticated job scheduling capabilities which are able to manage the workload running on a single Grid. We believe the major challenges are now around:

  • When and how to Grid enable an application. This is particularly true of financial services applications which can be called upon to provide interactive or very fast responses. This is because their are many jobs launched by traders out of Excel spreadsheets and also pricing of complex exotics in near real time. Traditional Grid technologies have been designed for much larger, longer running jobs and sacrifice responsiveness to achieve higher throughput. There are sometimes better ways to Grid enable, rather than just launching jobs at your favourite DRM. A standard Master-Worker pattern, for example, can be added to many distributed applications to better solve the problem in some cases.
  • Assessing the different emergent architectures such as Nvidia Cuda and FPGA based system, deciding when to use them and managing the deployment into critical production environments.
  • Managing persistence - traditional databases often cannot cope with the demand of hundreds of powerful Grid nodes or are just too slow for today's low latency requirements. There are several options for solving this including memory caches, data grids and SSD storage. We can can help determine which one to use where.
  • How to drive up Grid efficiency. We often see "Grid hugging" in large Banks, with as many as 80 independent Grids allocated on a per application basis. Whilst every one knows that different priorities can be set to provide some form of control over a shared Grid environment, Banks generally require higher levels of control. Instead, we propose a model where each business can retain a logical Grid with it's own DRM that they control, but a separate resource management scheme dynamically changes the capacities of each Grid based on pre-defined policies

Conference Presentations

Richard is a frequently invited speaker at Grid computing conferences, these have included: