Provably space-efficient parallel functional programming

Because of its many desirable properties, such as ability to control effects and thus potentially disastrous race conditions, functional programming offers a viable approach modern multicore computers. Over the past decade several parallel languages, typically based on dialects ML Haskell, have been developed. These however, traditionally underperformed procedural languages (such C Java). The primary reason for this is their hunger memory, which only grows with parallelism, causing traditional memory management techniques buckle under increased demand memory. Recent work opened new angle attack problem by identifying property determinacy-race-free programs, called disentanglement, limits knowledge concurrent computations about each other’s allocations. has showed some promise in delivering good time scalability. In paper, we present provably space-efficient automatic allowing both pure imperative programs where may be destructively updated. We prove that program sequential live R * , any P -processor garbage-collected run requires at most O ( · ) also bound W + executions, accounting cost garbage collection. To achieve these results, integrate thread scheduling management. idea coordinate allocation collection decisions so processor can allocate without synchronization independently collect portion consulting policy, formulate. policy fully distributed does not require communicating other processors. show practical implementing it an extension MPL compiler Parallel ML. Our experimental results confirm our theoretical bounds perform scale well.