Actors vs Shared Memory: two models at work on Big Data application frameworks

TL;DR

This paper compares shared memory and actor concurrency models in Big Data applications through conceptual analysis and experimental evaluation on MapReduce and BSP frameworks using Akka Cluster and Managed X10.

Contribution

It provides a comparative analysis of the two models in Big Data contexts, combining conceptual insights with experimental results on real platforms.

Findings

Actor model offers better scalability in Big Data scenarios.

Shared memory model simplifies development but has limitations in distributed environments.

Experimental results highlight performance differences between models.

Abstract

This work aims at analyzing how two different concurrency models, namely the shared memory model and the actor model, can influence the development of applications that manage huge masses of data, distinctive of Big Data applications. The paper compares the two models by analyzing a couple of concrete projects based on the MapReduce and Bulk Synchronous Parallel algorithmic schemes. Both projects are doubly implemented on two concrete platforms: Akka Cluster and Managed X10. The result is both a conceptual comparison of models in the Big Data Analytics scenario, and an experimental analysis based on concrete executions on a cluster platform.