Pierre Kuonen

SmartGRID: A Fully Decentralized Grid Scheduling Framework Supported by Swarm Intelligence

Resource management and scheduling has proven to be one of the key topics for grid computing. Nowadays, the resource management field is subdivided into low-level and high-level approaches. While low-level resource management systems normally concern the scheduling activities within a single virtual organization, high-level schedulers focus on the large scale resources utilization with unstable resource availability, low reliability networks, multi-policies, multi-administrative domains, etc. In this paper, we propose a decentralized framework named SmartGRID to tackle high-level grid resource management and scheduling. Within the SmartGRID framework, swarm intelligence algorithms are used for resource discovery and monitoring, standard protocols and schemes are adopted for scheduler interoperability, and an embedded plugin mechanism is provided to utilize multi-type external scheduling strategies. With a clearly decoupled layered architecture, SmartGRID has been designed to be a generic and modular environment to support intelligent and interoperable grid resource management upon a volatile, dynamics, and heterogeneous grid computing infrastructure.

Towards an Integrated Vision across Inter-cooperative Grid Virtual Organizations

Much work has been done to exploit the benefit brought by allowing job execution on distributed computational resources. Nodes are typically able to share jobs only within the same virtual organization, which is inherently bounded by various reasons such as the adopted information system or other agreed constraints. The problem raised by such limitation is thus related to finding a way to enable interoperation between nodes from different virtual organizations. n nWe introduce a novel technique for integrating visions from both resource users and providers, allowing to serve multiple virtual organizations as a whole. By means of snapshot data stored within each grid node, such as processing and interacting history, we propose a demand-centered heuristic scheduling approach named Critical Friend Community (CFC). To this end, a set of simplified community scheduling targeted algorithms and processing workflows are described. A prototype of our scheduling approach is being implemented within the SmartGRID project.

Using Metadata Snapshots for Extending Ant-Based Resource Discovery Service in Inter-cooperative Grid Communities

Much work is under way within the resource management community on issues associated with grid scheduling upon dynamically discovered information. In this paper we tackle the problem by exploiting a bio-inspired resource discovery mechanism, where information is provided by ant-based lightweight mobile agents traveling across a grid network and collecting data from each visited node. We start by providing the current state of the adopted grid scheduler, which is the result of an existing collaborative project named SmartGRID, and its underlying architecture constructed by ant-based mobile agents. We consider the problem of discovering resources in specific grid communities, which are bounded due to different shared community policies, such as diverse ant colonies, different resource discovery approaches, or other issues. Several issues have been raised during the design and implementation of such infrastructure. A notable issue, namely how grid schedulers from various bounded grid communities could be used in a manner which would extend current SmartGRID functionality is identified. Our shared view is that by utilizing already discovered and stored grid nodes’s metadata snapshots in the first instance we can facilitate a more convenient and efficient resource discovery operation next time. With this in mind, our paper goes on describing our shared vision with regard to this extended functionality as well as discussing the new conceptual basis and its model architecture.

A new approach to solve angular dispersion of discrete ray launching for urban scenarios

Ray-based methods such as ray tracing and ray launching have been increasingly used in radio wave propagation modelling. Ray tracing is used for point-to-point multipath prediction (for few receivers) while ray launching, being more adaptable, is more suitable for multi-point prediction. However, ray launching suffers from angular dispersion which causes rays to miss pixels when the distance from the emitter increases. Several solutions such as beam tracing or ray splitting have been proposed to resolve this, but this paper presents a new approach, which is suitable for discrete ray launching, to avoid the problem. Results show that by this approach, discrete ray launching is suitable for radio wave propagation modelling. Significant speedups are observed compared to traditional ray-based models via parallelization techniques such as multi-threading and distributed computing. Complex channel characteristics due to multipaths in the urban environment can be obtained via this method.

MaGate Simulator: A Simulation Environment for a Decentralized Grid Scheduler

This paper presents a simulator for of a decentralized modular grid scheduler named MaGate. MaGates design emphasizes scheduler interoperability by providing intelligent scheduling serving the grid community as a whole. Each MaGate scheduler instance is able to deal with dynamic scheduling conditions, with continuously arriving grid jobs. Received jobs are either allocated on local resources, or delegated to other MaGates for remote execution. The proposed MaGate simulator is based on GridSim toolkit and Alea simulator, and abstracts the features and behaviors of complex fundamental grid elements, such as grid jobs, grid resources, and grid users. Simulation of scheduling tasks is supported by a grid network overlay simulator executing distributed ant-based swarm intelligence algorithms to provide services such as group communication and resource discovery. For evaluation, a comparison of behaviors of different collaborative policies among a community of MaGates is provided. Results support the use of the proposed approach as a functional ready grid scheduler simulator.

Advancing Inter-cloud Resource Discovery Based on Past Service Experiences of Transient Resource Clustering

Resource discovery in large-scale distributed systems is a challenging issue, mainly due to the traditional centralized topologies in which the nodes of these systems are typically organized. This unified approach has been proven to be effectual for cluster-based grids and clouds but questionable for large scale, heterogeneous and dynamically interoperable e-infrastructure such as grids and inter-clouds. The latter e-infrastructures are less tolerant with regards to scalability, elasticity and flexibility. In this work, we explore the decentralized distributed search protocols that transform nodes of large size distributed systems to act as both clients and servers. Specifically, we extend the use of a job profile specification that is generated during the user job submission process. The exploitation of the job profile will allow us to orchestrate and group different user submissions into transient inter-cloud brokering groups that represent a temporary resource cluster according to the current service submission characteristics. We suggest the clustering submissions in an inter-cloud system with the view of forming notional short-term grouped nodes that may reform over the time. In this way, the resource discovery process is becoming dynamically, that is to say, a transient group of nodes is required advancing a single request based on current time, opposed to multiple requests as currently happens. We further propose an architecture that implies nodes orchestration based on previous resource requests as well as we model a service meta-registry for inter-cloud systems to store relevant information to service submission.

Community-Aware Scheduling Protocol for Grids

Much work has been done to exploit the effectiveness and efficiency of job scheduling upon distributed computational––resources. With regard to existing resource topology and administrative constraints, scheduling approaches are designed for different hierarchic layers, for example, scheduling for job queues of local resource management systems (local scheduling), and scheduling for job queues of high level schedulers (also known as meta-schedulers or grid schedulers). Such scheduling approaches mainly focus on optimizing job queues of the hosting nodes, which are interconnected with computational resources directly or indirectly. In the real world (or in a community-based grid), a grid is comprised of nodes with different computing power and scheduling preferences, which in turn, raise a notable opportunity that is to exploit and optimize the process of job sharing between reachable grid nodes via improving the job allocation and efficiency ratio. In our work, we introduce a novel scheduling protocol which dedicates to disseminate scheduling events happened on each involving node to as many candidate nodes as possible. By means of the proposed protocol, scheduling process of each received job consists of several phases with awareness of grid volatility, and dynamic scheduling and rescheduling is allowed as long as the job execution has not started yet. To this end, a set of concerning algorithms and processing steps are described. A prototype of our scheduling approach is being implemented within the SmartGRID project.

A Performance Evaluation of a Grid-Enabled Object-Oriented Parallel Outdoor Ray Launching for Wireless Network Coverage Prediction

The use of parallel technologies to solve complex scientific problems has gained increased popularity. The ray optical methods are deterministic propagation approaches that are based on geometrically searching paths between emitter and receivers. They offer higher accuracy than empirical models, but suffer from slow calculations on exhausted rays that have to be searched. In this paper, an object-oriented scheme based on POP-C++ for parallel objects to accelerate outdoor ray launching is described. Performance evaluation is presented to show that this parallel scheme is promising in outdoor wireless propagation modeling. The possibility of running this model in the distributed grid environment is also discussed. Results have shown the great potential of using such a parallel model to predict accurate outdoor wireless propagation scenarios within a short time.

A novel dynamic Q-learning-based scheduler technique for LTE-advanced technologies using neural networks

The tradeoff concept between system capacity and user fairness attracts a big interest in LTE-Advanced resource allocation strategies. By using static threshold values for throughput or fairness, regardless the network conditions, makes the scheduler to be inflexible when different tradeoff levels are required by the system. This paper proposes a novel dynamic neural Q-learning-based scheduling technique that achieves a flexible throughput-fairness tradeoff by offering optimal solutions according to the Channel Quality Indicator (CQI) for different classes of users. The Q-learning algorithm is used to adopt different policies of scheduling rules, at each Transmission Time Interval (TTI). The novel scheduling technique makes use of neural networks in order to estimate proper scheduling rules for different states which have not been explored yet. Simulation results indicate that the novel proposed method outperforms the existing scheduling techniques by maximizing the system throughput when different levels of fairness are required. Moreover, the system achieves a desired throughput-fairness tradeoff and an overall satisfaction for different classes of users.

Multi Objective Resource Scheduling in LTE Networks Using Reinforcement Learning

The use of the intelligent packet scheduling process is absolutely necessary in order to make the radio resources usage more efficient in recent high-bit-rate demanding radio access technologies such as Long Term Evolution LTE. Packet scheduling procedure works with various dispatching rules with different behaviors. In the literature, the scheduling disciplines are applied for the entire transmission sessions and the scheduler performance strongly depends on the exploited discipline. The method proposed in this paper aims to discuss how a straightforward schedule can be provided within the transmission time interval TTI sub-frame using a mixture of dispatching disciplines per TTI instead of a single rule adopted across the whole transmission. This is to maximize the system throughput while assuring the best user fairness. This requires adopting a policy of how to mix the rules and a refinement procedure to call the best rule each time. Two scheduling policies are proposed for how to mix the rules including use of Q learning algorithm for refining the policies. Simulation results indicate that the proposed methods outperform the existing scheduling techniques by maximizing the system throughput without harming the user fairness performance.