In Q4 2015 the European Commission funded the POP project with the objective to set up a parallel performance analysis Centre of Excellence (CoE). The 2.5-year project is now finished, having completed more than 125 assessments of parallel application performance studies and proof of concept demonstrations on how to achieve higher application performance.
The project has refined and promoted a high-level analysis and modelling methodology. Performance reports were produced giving an external assessment that provides insight into the performance characteristics of codes developed by parallel application developers. A high-level model quantifies factors such as load balance, serialisation, transfer or computational efficiencies providing fundamental insight on how application performance scales and recommendations were made on most appropriate directions to refactor the code to improve the application performance.
Service orientation has been a corner stone of the Centre of Excellence activities. Customers from different scientific and technical domains benefited from this service. Researchers in public institutions (e.g. universities and research laboratories), industry and SMEs were targeted to demonstrate that efficient parallel computing is valuable in all scientific and technical sectors. Around 25% of our studies are for SMEs in our attempt to promote performance analysis in non-traditional sectors.
Improving awareness of the importance of understanding application performance and promoting programming best practices requires cultural change with important benefits in the long term. We believe that the POP project has contributed significantly in this direction. Many of the insights and suggestions provided have been very well received by the customers. In many cases, performance gains between 10%, or 10x or 20x have been achieved by the proof of concept service or through code optimisations by the developers themselves. In cases where not all POP recommendations can be implemented in the short term, they will have an impact in steering the activities of code developers in the coming months and years.
We have the satisfaction of seeing the effort of our analysts has been very well received by our customers. The customer advocacy activities within the project allow us to closely monitor how customers perceive our services and has been an important tool in improving our procedures.
The POP CoE will continue its activities during the next months with limited resources. We expect to be able to continue them in the future under different programmes, supporting the improvement of productivity and overall efficiency in how research and industry sustainably use the available computing resources. Whilst the project has come to a completion, its impact in different dimensions will continue.
-- Jesus Labarta (BSC), POP Project Coordinator