The argument presented here is not that users require their COBOL applications to run across thousands of processors, or that such applications would generally benefit from such systems. The argument is that now and for the foreseeable future the investment in legacy systems and the lack of a performance-enhancing migration route effectively limits the spread of HPC into the commercial sector.
It appears the only cost-effective way to address these issues is to provide automatic parallelisation of legacy COBOL applications to exploit HPC. Ultimately these improvements are likely to alleviate much of the need for optimisation in software maintenance thus allowing programmers to focus on improved decision support and information management.
It is also important to recognise that new commercial applications are constantly being developed and new COBOL standards are emerging. It is important to further note that many fourth-generation tools generate COBOL code.
The COBOL-97 standard, incorporates objects. Partitioning an application into encapsulated objects should give a natural basis for decomposition of an application for concurrent execution. Initial analysis suggests that MICROFOCUS, with OBJECT COBOL, have adopted something similar to the SMALLTALK Object model. This means that classes have an active and singular role at run-time which makes partitioning more difficult. From the perspective of parallel implementation, initial analysis suggests that OBJECT COBOL needs some re-design to make it distributable and so enable the natural partitioning for concurrency.
The investigation discussed here hopes to proceed with the development of the outlined techniques and show automatic performance improvement for COBOL legacy systems using HPC. In addition, the demonstration of technologies and approaches that show the performance-enhancing capability of parallel HPC systems should increase the commercial sector's acceptance of HPC.