Fine-grained parallelism

The domain decomposition algorithm described in Section 4.4.1 can be parallelized in a number of different ways. The MD algorithm contains opportunities for independent operations on several different levels. In principle, the interaction on each particle can be calculated independently of all the others and the same goes for time-integration. For an in-depth discussion of these issues we refer to the review by Fincham [97,98]. This fine-grain parallelism is not really used in... 

This work is part of an ongoing effort to incorporate fine-grained parallelism models and constructs into existing programming languages, such as CPLEX (for C Parallel Language Extensions) [1], the C-h- "Technical Specification for C-h- Extensions for Parallelism" [2], or, the topic of this work, the tasklet model for Ada [3,4,5]. 

The current proposal to extend Ada with a fine-grained parallelism model is based on the notion of tasklets, which are non-schedulable computation units (similar to Cilk [6] or OpenMP [7] tasks). However, in contrast to the C and C-h- work, the principle behind this model is that the specification of parallelism is an abstraction that is not fully controlled by the programmer. Instead, parallelism is a notion that is under the control of the compiler and the run-time. The programmer uses special syntax to indicate where parallelism opportunities occur in the code, whilst the compiler and runtime co-operate to provide parallel execution, when possible. 

The structure of the paper is as follows. Section 2 provides a short summary of the previously proposed tasklet model for fine-grained parallelization of Ada code. Then, Section 3 proposes the underlying executing model for tasklets, whilst Section 4 shows how this model can be used in real-time systems and mentions some currently open issues. Finally, Section 5 provides some conclusions. 

Although no implementation exists of the complete proposal, parallelism only makes sense to provide faster computation. This work brings to the Ada world models which are widely used in other fine-grained parallelization approaches, where for the general case efficient solutions exist. For the case of real-time systems, addressing parallelism is still a challenge, and more research and experimentation is needed. 

---