arXiv subjects

Maria Kihl

Publications and source records attributed to Maria Kihl.

14 recordsLinked to original sources

arXiv2023

Resilient Cloud Control System: Realizing resilient cloud-based optimal control for cyber-physical systems

The transformation to smart factories and the automation of mobile robotics is partly driven by a growing availability of ubiquitous cloud technologies. In cyber-physical systems, such as control systems, critical parts can be migrated to a cloud for offloading, enabling collaborative processes, improved performance, and life-cycle management. Despite the performance uncertainty in a cloud and the intermediate networks, presently, even cloud native function services are being investigated for supporting critical applications that are sensitive to time-varying execution and communication delays. In this paper, we introduce, implement, and empirically evaluate an architecture that successfully allows predictive controllers to take advantage of cloud native technology. Our solution relies on continuously adapting the control system to the present Quality of Service of the cloud and the intermediate network. As our results show, this allows a control system to survive interruptions, noisy neighbors, and time-variant resource availability. Without the proposed solution, the control system will fail due to resource constraints and insufficient response times. Further, we also show a system that can seamlessly switch between clouds and that multiple controllers using shared resources consequentially self-adapt so that no controller fails its objective.

cs.NI↗

arXiv2022

Distributed Robust Geocast Multicast Routing for Inter-Vehicle Communication

Numerous protocols for geocast have been proposed in literature. It has been shown that explicit route setup approaches perform poorly with VANETs due to limited route lifetime and frequent network fragmentation. The broadcast based approaches have considerable redundancy and add significantly to the overhead of the protocol. A completely distributed and robust geocast approach is presented in this paper, that is resilient to frequent topology changes and network fragmentation. A distance-based backoff algorithm is used to reduce the number of hops and a novel mechanism to reduce redundant broadcasts is introduced. The performance of the proposed protocol is evaluated for various scenarios and compared with simple flooding and a protocol based on explicit route setup.

cs.NI↗

arXiv2021

FedApp: a Research Sandbox for Application Orchestration in Federated Clouds using OpenStack

Multi-cluster federation is envisioned to be the next-generation cloud infrastructure, where it will play a vital part in the realization of concepts such as edge and fog computing. Orchestrating applications in federated environments poses new challenges to well-known research problems in various fields, such as load-balancing, auto-scaling, resource allocation and service migration. However, as access to real multi-cluster infrastructure is limited, a test-bed that provides similar characteristics to a real system is in demand. To enable researchers in associated fields to quickly setup experiments in a federated cloud environment, we have created the open-source sandbox FedApp that simplifies the process of deploying multiple virtual clusters in an OpenStack environment with the possibility of adding realistic network characteristics between sites. Each cluster comes deployed with the open-source and production-grade container orchestrator Kubernetes, complete with federation-wide monitoring using Prometheus/Grafana and simplified inter-cluster microservice communication using Istio.

cs.DC↗

arXiv2021

Towards Soft Circuit Breaking in Service Meshes via Application-agnostic Caching

Service meshes factor out code dealing with inter-micro-service communication, such as circuit breaking. Circuit breaking actuation is currently limited to an "on/off" switch, i.e., a tripped circuit breaker will return an application-level error indicating service unavailability to the calling micro-service. This paper proposes a soft circuit breaker actuator, which returns cached data instead of an error. The overall resilience of a cloud application is improved if constituent micro-services return stale data, instead of no data at all. While caching is widely employed for serving web service traffic, its usage in inter-micro-service communication is lacking. Micro-services responses are highly dynamic, which requires carefully choosing adaptive time-to-life caching algorithms. We evaluate our approach through two experiments. First, we quantify the trade-off between traffic reduction and data staleness using a purpose-build service, thereby identifying algorithm configurations that keep data staleness at about 3% or less while reducing network load by up to 30%. Second, we quantify the network load reduction with the micro-service benchmark by Google Cloud called Hipster Shop. Our approach results in caching of about 80% of requests. Results show the feasibility and efficiency of our approach, which encourages implementing caching as a circuit breaking actuator in service meshes.

cs.NI↗

arXiv2020

Synchronization in Digital Twins for Industrial Control Systems

Digital twins, which are a new concept in industrial control systems (ICS), play a key role in realizing the vision of a smart factory, and they can have different effective use cases. With digital twins, we have virtual replicas of physical systems so that they precisely mirror the internal behavior of the physical systems. Hence, synchronization is necessary to keep the states of digital twins in sync with those of their physical counterparts. Otherwise, their behavior may be different from each other, and it can lead to wrong decisions about the system that can have catastrophic consequences. In this paper, we propose three different architectures for digital twins, and then by investigating their ability to follow the physical system's behavior, we will determine the best architecture, whose output has the lowest error compared with the physical system's output.

eess.SY↗

arXiv2020

Impact of etcd Deployment on Kubernetes, Istio, and Application Performance

By intrinsic necessity, Kubernetes is a complex platform. Its complexity makes conducting performance analysis in that environment fraught with difficulties and emergent behavior. Applications leveraging more "moving parts" such as the Istio service mesh makes the platform strictly more complex, not less. In this paper we study how underlying platform constitution and deployment affects application performance, specifically in Kubernetes-based environments. We alter platform constitution via use of native Kubernetes networking or Istio. Platform deployment is altered via etcd data storage location at two extremes on the performance spectrum: network disk and RAM disk. Our results show that etcd performance has a large impact on that of Kubernetes and its ability to perform orchestration actions, and thereby indirectly on the performance of the application. The implication is that systems researchers conducting performance evaluations cannot just consider their specific application as being under test, but must also take the underlying Kubernetes platform into account. To conduct experiments of scientific rigor, we developed an experiment framework for conducting repeatable and reproducible experiments. Our framework and resulting data set are openly available for the research community to build upon and reason about.

cs.NI↗

arXiv2019

An assisting Model Predictive Controller approach to Control over the Cloud

In this paper we develop a computational offloading strategy with graceful degradation for executing Model Predictive Control using the cloud. Backed up by previous work we simulate the control of a cyber-physical-system at high frequency and illustrate how the system can be improved using the edge while keeping the computational burden low.

eess.SY↗

arXiv2018

Towards Mission-Critical Control at the Edge and Over 5G

With the emergence of industrial IoT and cloud computing, and the advent of 5G and edge clouds, there are ambitious expectations on elasticity, economies of scale, and fast time to market for demanding use cases in the next generation of ICT networks. Responsiveness and reliability of wireless communication links and services in the cloud are set to improve significantly as the concept of edge clouds is becoming more prevalent. To enable industrial uptake we must provide cloud capacity in the networks but also a sufficient level of simplicity and self-sustainability in the software platforms. In this paper, we present a research test-bed built to study mission-critical control over the distributed edge cloud. We evaluate system properties using a conventional control application in the form of a Model Predictive Controller. Our cloud platform provides the means to continuously operate our mission-critical application while seamlessly relocating computations across geographically dispersed compute nodes. Through our use of 5G wireless radio, we allow for mobility and reliably provide compute resources with low latency, at the edge. The primary contribution of this paper is a state-of-the art, fully operational test-bed showing the potential for merged IoT, 5G, and cloud. We also provide an evaluation of the system while operating a mission-critical application and provide an outlook on a novel research direction.

eess.SY↗

arXiv2017

Spot Pricing in the Cloud Ecosystem: A Comparative Investigation

Background: Spot pricing is considered as a significant supplement for building a full-fledged market economy for the Cloud ecosystem. However, it seems that both providers and consumers are still hesitating to enter the Cloud spot market. The relevant academic community also has conflicting opinions about Cloud spot pricing in terms of revenue generation. Aim: This work aims to systematically identify, assess, synthesize and report the published evidence in favor of or against spot-price scheme compared with fixed-price scheme of Cloud computing, so as to help relieve the aforementioned conflict. Method: We employed the systematic literature review (SLR) method to collect and investigate the empirical studies of Cloud spot pricing indexed by major electronic libraries. Results: This SLR identified 61 primary studies that either delivered discussions or conducted experiments to perform comparison between spot pricing and fixed pricing in the Cloud domain. The reported benefits and limitations were summarized to facilitate cost-benefit analysis of being a Cloud spot pricing player, while four types of theories were distinguished to help both researchers and practitioners better understand the Cloud spot market. Conclusions: This SLR shows that the academic community strongly advocates the emerging Cloud spot market. Although there is still a lack of practical and easily deployable market-driven mechanisms, the overall findings of our work indicate that spot pricing plays a promising role in the sustainability of Cloud resource exploitation.

cs.DC↗

arXiv2017

Performance Overhead Comparison between Hypervisor and Container based Virtualization

The current virtualization solution in the Cloud widely relies on hypervisor-based technologies. Along with the recent popularity of Docker, the container-based virtualization starts receiving more attention for being a promising alternative. Since both of the virtualization solutions are not resource-free, their performance overheads would lead to negative impacts on the quality of Cloud services. To help fundamentally understand the performance difference between these two types of virtualization solutions, we use a physical machine with "just-enough" resource as a baseline to investigate the performance overhead of a standalone Docker container against a standalone virtual machine (VM). With findings contrary to the related work, our evaluation results show that the virtualization's performance overhead could vary not only on a feature-by-feature basis but also on a job-to-job basis. Although the container-based solution is undoubtedly lightweight, the hypervisor-based technology does not come with higher performance overhead in every case. For example, Docker containers particularly exhibit lower QoS in terms of storage transaction speed.

cs.DC↗

arXiv2017

On a Feedback Control-based Mechanism of Bidding for Cloud Spot Service

As a cost-effective option for Cloud consumers, spot service has been considered to be a significant supplement for building a full-fledged market economy for the Cloud ecosystem. However, unlike the static and straightforward way of trading on-demand and reserved Cloud services, the market-driven regulations of employing spot service could be too complicated for Cloud consumers to comprehensively understand. In particular, it would be both difficult and tedious for potential consumers to determine suitable bids from time to time. To reduce the complexity in applying spot resources, we propose to use a feedback control to help make bidding decisions. Based on an arccotangent-function-type system model, our novel bidding mechanism imitates fuzzy and intuitive human activities to refine and issue new bids according to previous errors. The validation is conducted by using Amazon's historical spot price trace to perform a set of simulations and comparisons. The result shows that the feedback control-based mechanism obtains a better trade-off between bidding rationality and success rate than the other five comparable strategies. Although this mechanism is only for black-box bidding (price prediction) at this current stage, it can be conveniently and gradually upgraded to take into account external constraints in the future.

cs.DC↗

arXiv2017

DoKnowMe: Towards a Domain Knowledge-driven Methodology for Performance Evaluation

Software engineering considers performance evaluation to be one of the key portions of software quality assurance. Unfortunately, there seems to be a lack of standard methodologies for performance evaluation even in the scope of experimental computer science. Inspired by the concept of "instantiation" in object-oriented programming, we distinguish the generic performance evaluation logic from the distributed and ad-hoc relevant studies, and develop an abstract evaluation methodology (by analogy of "class") we name Domain Knowledge-driven Methodology (DoKnowMe). By replacing five predefined domain-specific knowledge artefacts, DoKnowMe could be instantiated into specific methodologies (by analogy of "object") to guide evaluators in performance evaluation of different software and even computing systems. We also propose a generic validation framework with four indicators (i.e.~usefulness, feasibility, effectiveness and repeatability), and use it to validate DoKnowMe in the Cloud services evaluation domain. Given the positive and promising validation result, we plan to integrate more common evaluation strategies to improve DoKnowMe and further focus on the performance evaluation of Cloud autoscaler systems.

cs.DC↗

arXiv2017

Using a Predator-Prey Model to Explain Variations of Cloud Spot Price

The spot pricing scheme has been considered to be resource-efficient for providers and cost-effective for consumers in the Cloud market. Nevertheless, unlike the static and straightforward strategies of trading on-demand and reserved Cloud services, the market-driven mechanism for trading spot service would be complicated for both implementation and understanding. The largely invisible market activities and their complex interactions could especially make Cloud consumers hesitate to enter the spot market. To reduce the complexity in understanding the Cloud spot market, we decided to reveal the backend information behind spot price variations. Inspired by the methodology of reverse engineering, we developed a Predator-Prey model that can simulate the interactions between demand and resource based on the visible spot price traces. The simulation results have shown some basic regular patterns of market activities with respect to Amazon's spot instance type m3.large. Although the findings of this study need further validation by using practical data, our work essentially suggests a promising approach (i.e.~using a Predator-Prey model) to investigate spot market activities.

cs.DC↗

arXiv2017

A Survey on Modeling Energy Consumption of Cloud Applications: Deconstruction, State of the Art, and Trade-off Debates

Given the complexity and heterogeneity in Cloud computing scenarios, the modeling approach has widely been employed to investigate and analyze the energy consumption of Cloud applications, by abstracting real-world objects and processes that are difficult to observe or understand directly. It is clear that the abstraction sacrifices, and usually does not need, the complete reflection of the reality to be modeled. Consequently, current energy consumption models vary in terms of purposes, assumptions, application characteristics and environmental conditions, with possible overlaps between different research works. Therefore, it would be necessary and valuable to reveal the state-of-the-art of the existing modeling efforts, so as to weave different models together to facilitate comprehending and further investigating application energy consumption in the Cloud domain. By systematically selecting, assessing and synthesizing 76 relevant studies, we rationalized and organized over 30 energy consumption models with unified notations. To help investigate the existing models and facilitate future modeling work, we deconstructed the runtime execution and deployment environment of Cloud applications, and identified 18 environmental factors and 12 workload factors that would be influential on the energy consumption. In particular, there are complicated trade-offs and even debates when dealing with the combinational impacts of multiple factors.

cs.DC↗