Product designers and engineers are using an array of complex software which harness the power of parallel computing. They may be using CFD software, for example, to simulate a diesel engine combustion process and to do this they are finding they have a requirement for massive computer power to deliver the resolution needed within a sensible timeframe.
Is the answer to buy a supercomputer? This is a very significant investment, and one that requires a heavy compute workload to pay its way. Not to mention the power and cooling requirements, computer room and the specialist skills to manage such a system.
Plus, making a capital investment is fraught with danger for the uninitiated – high performance equipment ages quickly – and maintaining investment in new equipment can be a burden.
Is the answer to throw all eggs into the 'cloud basket' and rent supercomputing power? Well, a supercomputer accessible via the internet certainly enables businesses to access far larger compute power and only pay for what they actually use; they can use the service with confidence that all the equipment is fully up to date; and concerns over energy, cooling and space are transferred to the service provider.
Finding the cost-effective solution
But, it can be costly if used as the sole means for accessing compute power. Whilst compute cycle costs are relatively cheap, advanced commercial software licenses typically escalate in cost dramatically if the user wants to run on multiple CPU cores.
Our view of the market is that most organisations will accommodate the normal day-to-day workload with powerful workstations or small clusters which can handle maybe 80% of their typical modelling and simulation workloads.
The rest will be "burst" to an external supercomputer service that delivers first class performance and SLAs, at a manageable price.
This hybrid model is fine for users of open source software but can only flourish, however, if commercial software vendors begin to relax their traditional license models and embrace the "cloud", with licenses that do not penalise users for greater CPU use.
We're beginning to see some evidence of a shift in this area: for example Next Limit Technologies, who market an advanced CFD package called Xflow, is pioneering a token-based pricing system where there is minimal up-front cost and a genuine pay-per-use model. We expect other ISVs to follow, but perhaps not as quickly as users would like them to.
- Julian Fielden is managing director of OCF plc, providers of HPC, data management, data storage and data analytics. Drawing on twenty-years of business and accounting experience, Julian is responsible for the strategic direction of the business and its day-to-day operations.