Partitioning Granularity, Communication Overhead, and Adaptation in OS Services for Distributed Reconfigurable Systems on Chip

Distributed embedded systems are exposed to variable applications and requirements. This results in a dynamic and unpredictable changes in the availability of resources. Existing operating system implementations for embedded devices are struggling to address the challenges of efficiently utilizing modern target platforms with heterogeneous computational elements and variable resources. To cope with these challenges, operating system services can be partitioned into small blocks that can run on any computational element and use different resource combinations in the target platform. Although this partitioning allows flexibility and adaptation, it also introduces undesired inter-partition communication overhead. In this paper we discuss a lower bound for the partitioning granularity in order to provide a trade off between communication overhead and adaptation. In addition, we argue the usage of pipelining as an approach to reducing the communication overhead. As a proof of concept a case study is presented.