Using akka actors for managing iterations in multiscale applications

Abstract

In this work we investigate possibilities given by the modern technologies as Akka actor platform and Docker toolkit for managing execution of multiscale applications. We present fine-grained (i.e. iteration) level approach to this problem which is rather difficult to address in standard HPC environments with coarse-grained queuing systems. We propose the solution that uses Docker containers as wrappers for computational modules that are components of multiscale applications. This apprach allows for a fine-grained assignment of resources to such modules. We also investigate possibilities of Akka actor framework for building distributed steering systems for such applications with the REST protocol offered by the Spray toolkit. Results of an experiment validation shows that thanks to Docker features, sheduling application on a fine graned iteration level is possible. They also show that using Akka dispatcher as a scheduler for computational tasks (iterations) is a promising approach; moreover the Akka and Spray toolkits do not introduce a significant overhead.