Comparing the usability of library vs. language approaches to task parallelism

Abstract

In this paper, we compare the usability of a library approach with a language approach to task parallelism. There are many practical advantages and disadvantages to both approaches. A key advantage of a library-based approach is that it can be deployed without requiring any change in the tool chain, including compilers and IDEs. However, the use of library APIs to express all aspects of task parallelism can lead to code that is hard to understand and modify. A key advantage of a language-based approach is that the intent of the programmer is easier to express and understand, both by other programmers and by program analysis tools. However, a language-based approach usually requires the standardization of new constructs and (possibly) of new keywords. In this paper, we compare the java.util.concurrent (j.u.c) library [14] from Java 7 and the Habanero-Java (HJ) [16] language, supported by our experiences in teaching both models at Rice University.