Search

The article discusses emerging non-von Neumann computer architectures and their integration in the computing continuum for supporting modern distributed applications, including artificial intelligence, big data, and scientific computing. It provides a detailed summary of available and emerging non-von Neumann architectures, which range from power-efficient single-board accelerators to quantum and neuromorphic computers. Furthermore, it explores their potential benefits for revolutionizing data processing and analysis in various societal, science, and industry fields. The article provides a detailed analysis of the most widely used class of distributed applications and discusses the difficulties in their execution over the computing continuum, including communication, interoperability, orchestration, and sustainability issues.

The article discusses the emerging non-von Neumann computer architectures and their integration in the computing continuum for supporting modern distributed applications, including artificial intelligence, big data, and scientific computing. It provides a detailed summary of the available and emerging non-von Neumann architectures, which range from power-efficient single-board accelerators to quantum and neuromorphic computers. Furthermore, it explores their potential benefits for revolutionizing data processing and analysis in various societal, science, and industry fields. The paper provides a detailed analysis of the most widely used class of distributed applications and discusses the difficulties in their execution over the computing continuum, including communication, interoperability, orchestration, and sustainability issues.

The article discusses the emerging non-von Neumann computer architectures and their integration in the computing continuum for supporting modern distributed applications, including artificial intelligence, big data, and scientific computing. It provides a detailed summary of the available and emerging non-von Neumann architectures, which range from power-efficient single-board accelerators to quantum and neuromorphic computers. Furthermore, it explores their potential benefits for revolutionizing data processing and analysis in various societal, science, and industry fields. The paper provides a detailed analysis of the most widely used class of distributed applications and discusses the difficulties in their execution over the computing continuum, including communication, interoperability, orchestration, and sustainability issues.

Although serverless computing generally involves executing short-lived "functions", the increasing migration to this computing paradigm requires careful consideration of energy and power requirements. Serverless computing is also viewed as an economically-driven computational approach, often influenced by the cost of computation, as users are charged for per-sub-second use of computational resources rather than the coarse-grained charging that is common with virtual machines and containers. To ensure that the startup times of serverless functions do not discourage their use, resource providers need to keep these functions hot, often by passing in synthetic data. We describe the real power consumption characteristics of serverless, based on execution traces reported in the literature, and describe potential strategies (some adopted from existing VM and container-based approaches) that can be used to reduce the energy overheads of serverless execution. Our analysis is, purposefully, biased towards the use of machine learning workloads as: (i) such workloads are increasingly being used widely across different applications; (ii) functions that implement machine learning algorithms can range in complexity from long-running (deep learning) vs. short-running (inference only), enabling us to consider serverless across a variety of possible execution behaviours. The general findings are also easily translatable to other domains.

As Multi-Robot Systems (MRS) become more affordable and computing capabilities grow, they provide significant advantages for complex applications such as environmental monitoring, underwater inspections, or space exploration. However, accounting for potential communication loss or the unavailability of communication infrastructures in these application domains remains an open problem. Much of the applicable MRS research assumes that the system can sustain communication through proximity regulations and formation control or by devising a framework for separating and adhering to a predetermined plan for extended periods of disconnection. The latter technique enables an MRS to be more efficient, but breakdowns and environmental uncertainties can have a domino effect throughout the system, particularly when the mission goal is intricate or time-sensitive. To deal with this problem, our proposed framework has two main phases: i) a centralized planner to allocate mission tasks by rewarding intermittent rendezvous between robots to mitigate the effects of the unforeseen events during mission execution, and ii) a decentralized replanning scheme leveraging epistemic planning to formalize belief propagation and a Monte Carlo tree search for policy optimization given distributed rational belief updates. The proposed framework outperforms a baseline heuristic and is validated using simulations and experiments with aerial vehicles.

The article discusses the emerging non-von Neumann computer architectures and their integration in the computing continuum for supporting modern distributed applications, including artificial intelligence, big data, and scientific computing. It provides a detailed summary of the available and emerging non-von Neumann architectures, which range from power-efficient single-board accelerators to quantum and neuromorphic computers. Furthermore, it explores their potential benefits for revolutionizing data processing and analysis in various societal, science, and industry fields. The paper provides a detailed analysis of the most widely used class of distributed applications and discusses the difficulties in their execution over the computing continuum, including communication, interoperability, orchestration, and sustainability issues.

The article discusses the emerging non-von Neumann computer architectures and their integration in the computing continuum for supporting modern distributed applications, including artificial intelligence, big data, and scientific computing. It provides a detailed summary of the available and emerging non-von Neumann architectures, which range from power-efficient single-board accelerators to quantum and neuromorphic computers. Furthermore, it explores their potential benefits for revolutionizing data processing and analysis in various societal, science, and industry fields. The paper provides a detailed analysis of the most widely used class of distributed applications and discusses the difficulties in their execution over the computing continuum, including communication, interoperability, orchestration, and sustainability issues.

The article discusses the emerging non-von Neumann computer architectures and their integration in the computing continuum for supporting modern distributed applications, including artificial intelligence, big data, and scientific computing. It provides a detailed summary of the available and emerging non-von Neumann architectures, which range from power-efficient single-board accelerators to quantum and neuromorphic computers. Furthermore, it explores their potential benefits for revolutionizing data processing and analysis in various societal, science, and industry fields. The paper provides a detailed analysis of the most widely used class of distributed applications and discusses the difficulties in their execution over the computing continuum, including communication, interoperability, orchestration, and sustainability issues.

Front Matter, Table of Contents, Preface, Conference Organization

LIPIcs, Volume 262, ECRTS 2023, Complete Volume

Edge computing promises lower processing latencies and better privacy control than cloud computing for task offloading as edge devices are positioned closer to users. Realizing this promise depends on building strong theoretical and engineering foundations of computing based on an edge continuum connecting edge to other resources. In the SPEC-RG Cloud Group, we conducted a systematic study of computing models for task offloading and found that these models have many shared characteristics. Despite these commonalities, no systematic model or architecture for task offloading currently exists. In this paper, we address this need by proposing a reference architecture for task offloading in the edge continuum and synthesize its components using well-understood abstractions, services, and resources from cloud computing. We provide domain-specific architectures for deep learning and industrial IoT and show how this unified computing model opens up application development as developers are no longer limited to the single set of constraints posed by current isolated computing models. Additionally, we demonstrate the utility of the architecture by designing a deployment and benchmarking framework for edge continuum applications and investigate the performance of various edge continuum deployments. The framework allows for fine-grained discovery of the edge continuum deployment space, including the emulation of complex networks, all with minimal user input required. To enhance the performance analysis capabilities of the benchmark, we introduce an analytical first-order performance model that can be used to explore multiple application deployment scenarios such as local processing on endpoints or offloading between cloud or edge. The deployment and benchmarking framework is open-sourced and available at https://github.com/atlarge-research/continuum.

As the next generation of diverse workloads like autonomous driving and augmented/virtual reality evolves, computation is shifting from cloud-based services to the edge, leading to the emergence of a cloud-edge compute continuum. This continuum promises a wide spectrum of deployment opportunities for workloads that can leverage the strengths of cloud (scalable infrastructure, high reliability) and edge (energy efficient, low latencies). Despite its promises, the continuum has only been studied in silos of various computing models, thus lacking strong end-to-end theoretical and engineering foundations for computing and resource management across the continuum. Consequently, developers resort to ad hoc approaches to reason about performance and resource utilization of workloads in the continuum. In this work, we conduct a first-of-its-kind systematic study of various computing models, identify salient properties, and make a case to unify them under a compute continuum reference architecture. This architecture provides an end-to-end analysis framework for developers to reason about resource management, workload distribution, and performance analysis. We demonstrate the utility of the reference architecture by analyzing two popular continuum workloads, deep learning and industrial IoT. We have developed an accompanying deployment and benchmarking framework and first-order analytical model for quantitative reasoning of continuum workloads. The framework is open-sourced and available at https://github.com/atlarge-research/continuum., Comment: 14 pages, SPEC-RG technical report

In recent years, autonomous systems have become an important research area and application domain, with a significant impact on modern society. Such systems are characterized by different levels of autonomy and complex communication infrastructures that allow for collective decision-making strategies. There exist several publications that tackle ethical aspects in such systems, but mostly from the perspective of a single agent. In this paper we go one step further and discuss these ethical challenges from the perspective of an aggregate of autonomous systems capable of collective decision-making. In particular, in this paper, we propose the Caesar approach through which we model the collective ethical decision-making process of a group of actors—agents and humans, as well as define the building blocks for the agents participating in such a process, namely Caesar agents. Factors such as trust, security, safety, and privacy, which affect the degree to which a collective decision is ethical, are explicitly captured in Caesar. Finally, we argue that modeling the collective decision-making in Caesar provides support for accountability.

Robotics automation is a promising solution for the fabrication of structures made out of reinforced concrete. The reinforcement is often installed directly in the form and barby-bar. Using bigger pre-fabricated units (cages) may be beneficial for saving construction time and better labor safety. In this paper, we focus on the problem of automating the generation of a plan for the installation of rebars, given the digital twin of a desired reinforcement cage design, and of its basic components. More specifically, the plan describes the assembling order for the rebars such that (i) it is possible to fabricate the reinforcement cage by the robots, and (ii) the end product is the final reinforcement cage, ready for installation in the form for the concrete structure. In this paper, we propose an algorithm to automatically compute a feasible installation order for a generic rebar cage. The feasibility of the generated order is also case studied and simulated on a simplified rebar cage under the given assumptions.

Stream processing applications extract value from raw data through Directed Acyclic Graphs of data analysis tasks. Shared-nothing (SN) parallelism is the de-facto standard to scale stream processing applications. Given an application, SN parallelism ins9tantiates several copies of each analysis task, making each instance responsible for a dedicated portion of the overall analysis, and relies on dedicated queues to exchange data among connected instances. On the one hand, SN parallelism can scale the execution of applications both up and out since threads can run task instances within and across processes/nodes. On the other hand, its lack of sharing can cause unnecessary overheads and hinder the scaling up when threads operate on data that could be jointly accessed in shared memory. This trade-off motivated us in studying a way for stream processing applications to leverage shared memory and boost the scale up (before the scale out) while adhering to the widely-adopted and SN-based APIs for stream processing applications. We introduce STRETCH, a framework that maximizes the scale up and offers instantaneous elastic reconfigurations (without state transfer) for stream processing applications. We propose the concept of Virtual Shared-Nothing (VSN) parallelism and elasticity and provide formal definitions and correctness proofs for the semantics of the analysis tasks supported by STRETCH, showing they extend the ones found in common Stream Processing Engines. We also provide a fully implemented prototype and show that STRETCH's performance exceeds that of state-of-the-art frameworks such as Apache Flink and offers, to the best of our knowledge, unprecedented ultra-fast reconfigurations, taking less than 40 ms even when provisioning tens of new task instances.

Edge computing promises lower processing latencies and better privacy control than cloud computing for task offloading as edge devices are positioned closer to users. Realizing this promise depends on building strong theoretical and engineering foundations of computing based on an edge continuum connecting edge to other resources. In the SPEC-RG Cloud Group, we conducted a systematic study of computing models for task offloading and found that these models have many shared characteristics. Despite these commonalities, no systematic model or architecture for task offloading currently exists. In this paper, we address this need by proposing a reference architecture for task offloading in the edge continuum and synthesize its components using well-understood abstractions, services, and resources from cloud computing. We provide domain-specific architectures for deep learning and industrial IoT and show how this unified computing model opens up application development as developers are no longer limited to the single set of constraints posed by current isolated computing models. Additionally, we demonstrate the utility of the architecture by designing a deployment and benchmarking framework for edge continuum applications and investigate the performance of various edge continuum deployments. The framework allows for fine-grained discovery of the edge continuum deployment space, including the emulation of complex networks, all with minimal user input required. To enhance the performance analysis capabilities of the benchmark, we introduce an analytical first-order performance model that can be used to explore multiple application deployment scenarios such as local processing on endpoints or offloading between cloud or edge. The deployment and benchmarking framework is open-sourced and available at https://github.com/atlarge-research/continuum.

A traditional approach to realize self-adaptation in software engineering (SE) is by means of feedback loops. The goals of the system can be specified as formal properties that are verified against models of the system. On the other hand, control theory (CT) provides a well-established foundation for designing feedback loop systems and providing guarantees for essential properties, such as stability, settling time, and steady state error. Currently, it is an open question whether and how traditional SE approaches to self-adaptation consider properties from CT. Answering this question is challenging given the principle differences in representing properties in both fields. In this paper, we take a first step to answer this question. We follow a bottom up approach where we specify a control design (in Simulink) for a case inspired by Scuderia Ferrari (F1) and provide evidence for stability and safety. The design is then transferred into code (in C) that is further optimized. Next, we define properties that enable verifying whether the control properties still hold at code level. Then, we consolidate the solution by mapping the properties in both worlds using specification patterns as common language and we verify the correctness of this mapping. The mapping offers a reusable artifact to solve similar problems. Finally, we outline opportunities for future work, particularly to refine and extend the mapping and investigate how it can improve the engineering of self-adaptive systems for both SE and CT engineers. © 2021 IEEE.

Stream processing applications extract value from raw data through Directed Acyclic Graphs of data analysis tasks. Shared-nothing (SN) parallelism is the de-facto standard to scale stream processing applications. Given an application, SN parallelism instantiates several copies of each analysis task, making each instance responsible for a dedicated portion of the overall analysis, and relies on dedicated queues to exchange data among connected instances. On the one hand, SN parallelism can scale the execution of applications both up and out since threads can run task instances within and across processes/nodes. On the other hand, its lack of sharing can cause unnecessary overheads and hinder the scaling up when threads operate on data that could be jointly accessed in shared memory. This trade-off motivated us in studying a way for stream processing applications to leverage shared memory and boost the scale up (before the scale out) while adhering to the widely-adopted and SN-based APIs for stream processing applications. We introduce STRETCH, a framework that maximizes the scale up and offers instantaneous elastic reconfigurations (without state transfer) for stream processing applications. We propose the concept of Virtual Shared-Nothing (VSN) parallelism and elasticity and provide formal definitions and correctness proofs for the semantics of the analysis tasks supported by STRETCH, showing they extend the ones found in common Stream Processing Engines. We also provide a fully implemented prototype and show that STRETCH's performance exceeds that of state-of-the-art frameworks such as Apache Flink and offers, to the best of our knowledge, unprecedented ultra-fast reconfigurations, taking less than 40 ms even when provisioning tens of new task instances.

A traditional approach to realize self-adaptation in software engineering (SE) is by means of feedback loops. The goals of the system can be specified as formal properties that are verified against models of the system. On the other hand, control theory (CT) provides a well-established foundation for designing feedback loop systems and providing guarantees for essential properties, such as stability, settling time, and steady state error. Currently, it is an open question whether and how traditional SE approaches to self-adaptation consider properties from CT. Answering this question is challenging given the principle differences in representing properties in both fields. In this paper, we take a first step to answer this question. We follow a bottom up approach where we specify a control design (in Simulink) for a case inspired by Scuderia Ferrari (F1) and provide evidence for stability and safety. The design is then transferred into code (in C) that is further optimized. Next, we define properties that enable verifying whether the control properties still hold at code level. Then, we consolidate the solution by mapping the properties in both worlds using specification patterns as common language and we verify the correctness of this mapping. The mapping offers a reusable artifact to solve similar problems. Finally, we outline opportunities for future work, particularly to refine and extend the mapping and investigate how it can improve the engineering of self-adaptive systems for both SE and CT engineers.

Front Matter, Table of Contents, Preface, Conference Organization

Container-based virtualization has gained a significant importance in a deployment of software applications in cloud-based environments. The technology fully relies on operating system features and does not require a virtualization layer (hypervisor) that introduces a performance degradation. Container-based virtualization allows to co-locate multiple isolated containers on a single computation node as well as to decompose an application into multiple containers distributed among several hosts (e.g., in fog computing layer). Such a technology seems very promising in other domains as well, e.g., in industrial automation, automotive, and aviation industry where mixed criticality containerized applications from various vendors can be co-located on shared resources. However, such industrial domains often require real-time behavior (i.e, a capability to meet predefined deadlines). These capabilities are not fully supported by the container-based virtualization yet. In this work, we provide a systematic literature survey study that summarizes the effort of the research community on bringing real-time properties in container-based virtualization. We categorize existing work into main research areas and identify possible immature points of the technology.

In this work, we report on our recent article on "Methodological Principles for Reproducible Performance Evaluation in Cloud Computing" published in the IEEE Transactions on Software Engineering (TSE)[Pa19]. This publication is a result of a joint effort coordinated by the Cloud Research Working Group of SPEC (Standard Performance Evaluation Corporation).

Two of the main paradigms used to build adaptive software employ different types of properties to capture relevant aspects of the system's run-time behavior. On the one hand, control systems consider properties that concern static aspects like stability, as well as dynamic properties that capture the transient evolution of variables such as settling time. On the other hand, self-adaptive systems consider mostly non-functional properties that capture concerns such as performance, reliability, and cost. In general, it is not easy to reconcile these two types of properties or identify under which conditions they constitute a good fit to provide run-time guarantees. There is a need of identifying the key properties in the areas of control and self-adaptation, as well as of characterizing and mapping them to better understand how they relate and possibly complement each other. In this paper, we take a first step to tackle this problem by: (1) identifying a set of key properties in control theory, (2) illustrating the formalization of some of these properties employing temporal logic languages commonly used to engineer self-adaptive software systems, and (3) illustrating how to map key properties that characterize self-adaptive software systems into control properties, leveraging their formalization in temporal logics. We illustrate the different steps of the mapping on an exemplar case in the cloud computing domain and conclude with identifying open challenges in the area.

Container-based virtualization has gained a significant importance in a deployment of software applications in cloud-based environments. The technology fully relies on operating system features and does not require a virtualization layer (hypervisor) that introduces a performance degradation. Container-based virtualization allows to co-locate multiple isolated containers on a single computation node as well as to decompose an application into multiple containers distributed among several hosts (e.g., in fog computing layer). Such a technology seems very promising in other domains as well, e.g., in industrial automation, automotive, and aviation industry where mixed criticality containerized applications from various vendors can be co-located on shared resources. However, such industrial domains often require real-time behavior (i.e, a capability to meet predefined deadlines). These capabilities are not fully supported by the container-based virtualization yet. In this work, we provide a systematic literature survey study that summarizes the effort of the research community on bringing real-time properties in container-based virtualization. We categorize existing work into main research areas and identify possible immature points of the technology.

Front Matter, Table of Contents, Preface, Conference Organization

Streaming analysis is widely used in cloud as well as edge infrastructures. In these contexts, fine-grained application performance can be based on accurate modeling of streaming operators. This is especially beneficial for computationally expensive operators like adaptive stream joins that, being very sensitive to rate-varying data streams, would otherwise require costly frequent monitoring. We propose a dynamic model for the processing throughput and latency of adaptive stream joins that run with different parallelism degrees. The model is presented with progressive complexity, from a centralized non-deterministic up to a deterministic parallel stream join, describing how throughput and latency dynamics are influenced by various configuration parameters. The model is catalytic for understanding the behavior of stream joins against different system deployments, as we show with our model-based autoscaling methodology to change the parallelism degree of stream joins during the execution. Our thorough evaluation, for a broad spectrum of parameter, confirms the model can reliably predict throughput and latency metrics with a fairly high accuracy, with the median error in estimation ranging from approximately 0.1% to 6.5%, even for an overloaded system. Furthermore, we show that our model allows to efficiently control adaptive stream joins by estimating the needed resources solely based on the observed input load. In particular, we show it can be employed to enable efficient autoscaling, even when big changes in the input load happen frequently (in the realm of seconds)., Comment: part of the experimental campaign has been included in a different contribution

Two of the main paradigms used to build adaptive software employ different types of properties to capture relevant aspects of the system's run-time behavior. On the one hand, control systems consider properties that concern static aspects like stability, as well as dynamic properties that capture the transient evolution of variables such as settling time. On the other hand, self-adaptive systems consider mostly non-functional properties that capture concerns such as performance, reliability, and cost. In general, it is not easy to reconcile these two types of properties or identify under which conditions they constitute a good fit to provide run-time guarantees. There is a need of identifying the key properties in the areas of control and self-adaptation, as well as of characterizing and mapping them to better understand how they relate and possibly complement each other. In this paper, we take a first step to tackle this problem by: (1) identifying a set of key properties in control theory, (2) illustrating the formalization of some of these properties employing temporal logic languages commonly used to engineer self-adaptive software systems, and (3) illustrating how to map key properties that characterize self-adaptive software systems into control properties, leveraging their formalization in temporal logics. We illustrate the different steps of the mapping on an exemplar case in the cloud computing domain and conclude with identifying open challenges in the area.

The growing popularity of workflows in the cloud domain promoted the development of sophisticated autoscaling policies that allow automatic allocation and deallocation of resources. However, many state-of-the-art autoscaling policies for workflows are mostly plan-based or designed for batches (ensembles) of workflows. This reduces their flexibility when dealing with workloads of workflows, as the workloads are often subject to unpredictable resource demand fluctuations. Moreover, autoscaling in clouds almost always imposes budget constraints that should be satisfied. The budget-aware autoscalers for workflows usually require task runtime estimates to be provided beforehand, which is not always possible when dealing with workloads due to their dynamic nature. To address these issues, we propose a novel Performance-Feedback Autoscaler (PFA) that is budget-aware and does not require task runtime estimates for its operation. Instead, it uses the performance-feedback loop that monitors the average throughput on each resource type. We implement PFA in the popular Apache Airflow workflow management system, and compare the performance of our autoscaler with other two state-of-the-art autoscalers, and with the optimal solution obtained with the Mixed Integer Programming approach. Our results show that PFA outperforms other considered online autoscalers, as it effectively minimizes the average job slowdown by up to 47% while still satisfying the budget constraints. Moreover, PFA shows by up to 76% lower average runtime than the competitors., Comment: Technical Report

Elasticity is one of the main features of cloud computing allowing customers to scale their resources based on the workload. Many autoscalers have been proposed in the past decade to decide on behalf of cloud customers when and how to provision resources to a cloud application based on the workload utilizing cloud elasticity features. However, in prior work, when a new policy is proposed, it is seldom compared to the state-of-the-art, and is often compared only to static provisioning using a predefined quality of service target. This reduces the ability of cloud customers and of cloud operators to choose and deploy an autoscaling policy, as there is seldom enough analysis on the performance of the autoscalers in different operating conditions and with different applications. In our work, we conduct an experimental performance evaluation of autoscaling policies, using as application model workflows, a popular formalism for automating resource management for applications with well-defined yet complex structures. We present a detailed comparative study of general state-of-the-art autoscaling policies, along with two new workflow-specific policies. To understand the performance differences between the seven policies, we conduct various experiments and compare their performance in both pairwise and group comparisons. We report both individual and aggregated metrics. As many workflows have deadline requirements on the tasks, we study the effect of autoscaling on workflow deadlines. Additionally, we look into the effect of autoscaling on the accounted and hourly based charged costs, and we evaluate performance variability caused by the autoscaler selection for each group of workflow sizes. Our results highlight the trade-offs between the suggested policies, how they can impact meeting the deadlines, and how they perform in different operating conditions, thus enabling a better understanding of the current state-of-the-art.

The Mobile Cloud Network is an emerging cost and capacity heterogeneous distributed cloud topological paradigm that aims to remedy the application performance constraints imposed by centralised cloud infrastructures. A centralised cloud infrastructure and the adjoining Telecom network will struggle to accommodate the exploding amount of traffic generated by forthcoming highly interactive applications. Cost effectively managing a Mobile Cloud Network computing infrastructure while meeting individual application's performance goals is non-trivial and is at the core of our contribution. Due to the scale of a Mobile Cloud Network, a centralised approach is infeasible. Therefore, in this paper a distributed algorithm that addresses these challenges is presented. The presented approach works towards meeting individual application's performance objectives, constricting system-wide operational cost, and mitigating resource usage skewness. The presented distributed algorithm does so by iteratively and independently acting on the objectives of each component with a common heuristic objective function. Systematic evaluations reveal that the presented algorithm quickly converges and performs near optimal in terms of system-wide operational cost and application performance, and significantly outperforms similar naïve and random methods.

The Mobile Cloud Network is an emerging cost and capacity heterogeneous distributed cloud topological paradigm that aims to remedy the application performance constraints imposed by centralised cloud infrastructures. A centralised cloud infrastructure and the adjoining Telecom network will struggle to accommodate the exploding amount of traffic generated by forthcoming highly interactive applications. Cost effectively managing a Mobile Cloud Network computing infrastructure while meeting individual application's performance goals is non-trivial and is at the core of our contribution. Due to the scale of a Mobile Cloud Network, a centralised approach is infeasible. Therefore, in this paper a distributed algorithm that addresses these challenges is presented. The presented approach works towards meeting individual application's performance objectives, constricting system-wide operational cost, and mitigating resource usage skewness. The presented distributed algorithm does so by iteratively and independently acting on the objectives of each component with a common heuristic objective function. Systematic evaluations reveal that the presented algorithm quickly converges and performs near optimal in terms of system-wide operational cost and application performance, and significantly outperforms similar naïve and random methods.

Distributed control systems transform with the Industry 4.0 paradigm shift. A mesh-like, network-centric topologyreplaces the traditional controller-centered architecture, enforcing the interest of cloud-, fog-, and edge-computing, where lightweight container-based virtualization is a cornerstone. Kubernetes is a well-known container management system for container orchestration in cloud computing. It is gaining traction inedge- and fog-computing due to its elasticity and failure recovery properties. Orchestrator failure recovery can complement the manual replacement of a failed controller and, combined with controller redundancy, provide a pseudo-one-out-of-many redundancy. This paper investigates the failure recovery performance obtained from an out-of-the-box Kubernetes installation in a distributed control system scenario. We describe a Kubernetes based virtualized controller architecture and the software needed to setup a bare-metal cluster for control systems. Further, we deploy single and redundant configured containerized controllers based on an OPC UA compatible industry middleware software on the bare-metal cluster. The controllers expose variables with OPCUA PubSub. A script-based daemon introduces node failures, and a verification controller measures the downtime when using Kubernetes with an industry redundancy solution