Safe Parallel Programming in an Interpreted Language

Parallel programming is increasingly important with the advent of multicore processors. However, modern software is difficult to parallelize because of the high degree of modularization. It is unclear whether a piece of code is parallel if it calls other functions. Dynamic languages such as Ruby, Python, and Matlab represent modularization to the extreme. A program, also known as a script, requires interpretation for every statement, calling a whole host of functions from parsing to execution. Because of the complexity of the interpreter and the dynamic nature of script code, scripting languages are among the most difficult to parallelize. In this paper, safe parallelization means to parallelize a sequential program with the guarantee that the parallelized version produces the same result as the original sequential code. In other words, safety means sequential equivalence. There are generally three ways to ensure safety. Consider a sequential loop as the target program to parallelize. The first way is static. A compiler analyzes the loop and determines whether it is parallel. Static parallelization means that the loop is parallel in all executions. The second way is dynamic. Immediately before executing the loop, a run-time library determines whether the loop is parallel. The third way is speculative. The loop is run in parallel speculatively. If conflicts are detected in the parallel execution, the conflicting part is rolled back and re-run sequentially. This paper uses the third way and describes a system that speculatively parallelizes a Ruby script to ensure safety. In the past, most speculation systems were for automatic parallelization and were not user programmable [2, 5, 10]. Then a number of interfaces appeared, including safe future in Java [13], ordered transactions using TLS hardware [12], and possibly parallel region (PPR) in the BOP (behaviorbased parallelization) system [4, 6, 14], and multi-threaded transactions [9]. These are effectively hints of parallelism. Dependence hints were introduced in the BOP system as a way to suggest possible dependences between operations in different PPRs [7]. In this paper, we adapt the BOP interfaces for use in Ruby. 2. Safe Parallel Ruby