Fog resource selection using historical executions

Mostafa, Nour; Ridhawi, Ismaeel Al; Aloqaily, Moayad

doi:10.1109/fmec.2018.8364078

Cited by 21 publications

(9 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This means that before any neural network can be created, other placement algorithms or service placement policies have to be used to historically record those placement decisions to receive a viable training dataset. Mostafa, Ridhawi & Aloqaily (2018) also implement an artificial neural network in a simulated environment to make placement predictions based on historical placement data. The algorithms in our work could provide a basis for training data and eventually be substituted by ML models.…”

Section: Service Placementmentioning

confidence: 99%

See 1 more Smart Citation

FogFrame: a framework for IoT application execution in the fog

Skarlat¹,

Schulte²

2021

PeerJ Computer Science

View full text Add to dashboard Cite

Recently, a multitude of conceptual architectures and theoretical foundations for fog computing have been proposed. Despite this, there is still a lack of concrete frameworks to setup real-world fog landscapes. In this work, we design and implement the fog computing framework FogFrame—a system able to manage and monitor edge and cloud resources in fog landscapes and to execute Internet of Things (IoT) applications. FogFrame provides communication and interaction as well as application management within a fog landscape, namely, decentralized service placement, deployment and execution. For service placement, we formalize a system model, define an objective function and constraints, and solve the problem implementing a greedy algorithm and a genetic algorithm. The framework is evaluated with regard to Quality of Service parameters of IoT applications and the utilization of fog resources using a real-world operational testbed. The evaluation shows that the service placement is adapted according to the demand and the available resources in the fog landscape. The greedy placement leads to the maximum utilization of edge devices keeping at the edge as many services as possible, while the placement based on the genetic algorithm keeps devices from overloads by balancing between the cloud and edge. When comparing edge and cloud deployment, the service deployment time at the edge takes 14% of the deployment time in the cloud. If fog resources are utilized at maximum capacity, and a new application request arrives with the need of certain sensor equipment, service deployment becomes impossible, and the application needs to be delegated to other fog resources. The genetic algorithm allows to better accommodate new applications and keep the utilization of edge devices at about 50% CPU. During the experiments, the framework successfully reacts to runtime events: (i) services are recovered when devices disappear from the fog landscape; (ii) cloud resources and highly utilized devices are released by migrating services to new devices; (iii) and in case of overloads, services are migrated in order to release resources.

show abstract

Section: Service Placementmentioning

confidence: 99%

“… Mostafa, Ridhawi & Aloqaily (2018) also implement an artificial neural network in a simulated environment to make placement predictions based on historical placement data. The algorithms in our work could provide a basis for training data and eventually be substituted by ML models.…”

Section: Related Workmentioning

confidence: 99%

FogFrame: a framework for IoT application execution in the fog

Skarlat¹,

Schulte²

2021

PeerJ Computer Science

View full text Add to dashboard Cite

show abstract

“…Authors achieved an increment of 10% CPU usage and 20% RAM usage compared to static management systems without significantly reducing service quality. In [16], the authors propose a resource selection algorithm in fog computing (FResS) that allows automatic selection and allocation for IoT systems. The proposed model in the article maintains a repository of performance data in the form of execution records.…”

Section: B Algorithms For Resource Consumption Predictionmentioning

confidence: 99%

Provisioning Computational Resources for Cloud-Based e-Learning Platforms Using Deep Learning Techniques

et al. 2021

View full text Add to dashboard Cite

The use of e-learning technologies is growing even faster due to the existing conditions where virtual setups temporarily replace traditional classroom environments. Service infrastructure support for e-learning has moved to the cloud. For this reason, the efficient provisioning of resources for such platforms, which is achieved through prediction, is very relevant. The existing techniques for predicting the use of resources in the cloud are not designed with e-learning's specific requirements. This paper presents a neural-network-based model for predicting the usage of computational resources for e-learning platforms. This model consists of a series of interconnected neural networks used to predict values for variables of interest, such as Random Access Memory (RAM) usage and Central Processing Unit (CPU) usage. Using data collected from a high school real scenario, we analyzed and used it to train and validate our neuralnetwork-based model. This scenario consisted of a Moodle server deployed in a Google Virtual Machine with a configured course and its contents. Each student performed a series of activities while connected to it. Our proposed model achieves high accuracy. The obtained results are promising, paving the way towards the construction of software tools for provisioning computational resources on demand for e-learning platforms.

show abstract

“…For more details in regards to the TTP service mediation process, the reader may refer to [31]- [33]. Finally, for service discovery and selection, the reader may refer to [34]- [36].…”

Section: Problem and Solution Overviewmentioning

confidence: 99%