Conference paper

Right On Time Distributed Shared Memory

The demand for real-time data storage in distributed control systems (DCSs) is growing. Yet, providing real- time DCS guarantees is challenging, especially when more and more sensor and actuator devices are connected to industrial plants and message loss needs to be taken into account. In this paper, we investigate how to build a shared memory abstraction for DCSs as a first step towards implementing different shared storage systems in a DCS context. We first prove that, in the presence of host crashes and message losses, the necessary guarantees of such an abstraction are impossible to implement using a traditional approach that has no access to the internals of existing DCS services, e.g., a modular approach where algorithms are built on top of existing software blocks like failure detectors. We propose a white-box approach that utilizes messages of existing services in any DCS as the sole means of communication. More precisely, we present TapeWorm, an algorithm that attaches itself to the heartbeat messages of the failure detector component in DCSs. We prove that TapeWorm implements the desired shared memory guarantees for applications running on a DCS. We also analyze the performance of TapeWorm and we showcase ways of adapting TapeWorm to various application needs and workloads.

Related material