C++ library for parallel programming on a distributed system

Abstract

The goal of this work was to develop an efficient and convenient API for distributed parallel computing and implement it as a library. The library provides an interface to describe parallel computation on distributed systems in terms of tasks and their relations. A high level of task description allows to manage and maintain task execution on a cluster efficiently. The library handles distributed task dispatching, scheduling, basic I/O functionalities, and node communication. As a result, the programmer doesn’t have to worry about cluster management and can focus on algorithms. The idea of the library was inspired by the Intel Thread building blocks library for the shared memory systems.