ABSTRACT Background Federated learning on the edge allows the use of more computing capacity and more complex training models while providing higher accuracy and bandwidth savings. However, the deployment and management of edge cloud infrastructures is still challenging due to its highly distributed nature. Aims This paper presents the deployment of a real framework for federated learning on top of a real geo‐distributed edge cloud infrastructure. Methods The OpenNebula cloud platform is used to deploy an edge cloud infrastructure and a federated learning framework on top of it, as well as to manage and orchestrate the infrastructure. Results A global edge‐based infrastructure was deployed in 16 min for just $6 per hour, reducing execution time and increasing the accuracy of the trained model. Conclusion Results show the feasibility, performance, and cost efficiency of the solution, which provides an easy, fast, and inexpensive way to perform edge‐based federated learning.

We are pleased to present a special issue that focuses on the trends in the next generation of cloud computing. An obvious question that may arise in the mind of the reader is why another issue on clouds when there is plenty of discourse on the topic. Undoubtedly, the cloud computing landscape is rapidly changing to meet the challenges of emerging paradigms, such as Fog/Edge computing and the Internet‐of‐Things (IoT). These paradigms rely on services offered by the cloud but are demanding—need to scale for billions of heterogeneous devices and sensors while operating efficiently in real‐time.

Telecom companies and infrastructure providers are planning to deploy thousands of processing nodes at the edge of their networks to provide end users and companies with low-latency edge services for IoT, video streaming, gaming, and mobile applications. However, the efficient management of these massively distributed edge computing environments is still an open issue. In this article, we present a disaggregated cloud management model that would allow companies and institutions to deploy their own edge computing infrastructures and services by combining their on-premises cloud infrastructure (if applicable) with multiple highly-dispersed edge nodes leased from third-party providers. We compare this novel disaggregated model, with the more typical distributed one. Results show that the proposed model is suitable for managing small-to medium-sized edge computing infrastructures.

Summary This paper describes the GridWay metascheduler and exposes its latest and future developments, mainly related to interoperability and interoperation. GridWay enables large‐scale, reliable, and efficient sharing of computing resources over grid middleware. To favor interoperability, it shows a modular architecture based on drivers, which access middleware services for resource discovery and monitoring, job execution and management, and file transfer. This paper presents two new execution drivers for Basic Execution Service (BES) and Computing Resource Execution and Management (CREAM) services and introduces a remote BES interface for GridWay. This interface allows users to access GridWay's job metascheduling capabilities, using the BES implementation of GridSAM. Thus, GridWay now provides to end users more possibilities of interoperability and interoperation.Copyright © 2012 John Wiley & Sons, Ltd.

Summary The number of providers in the cloud computing market is increasing at a rapid pace. They offer a wide range of pricing schemes, different types of instance, or even different value‐added features to differ from other competitors, making the cloud market more complex. Cloud brokering enable users to choose the best cloud provider for their needs and avoid particular providers' lock in. Moreover, the use of multiple clouds offers several benefits such as scalability of services, improvement in fault tolerance, or cost reduction. However, users still find difficult the decision of where to deploy their resources as they have to handle too much information. The use of cloud brokering mechanisms is useful to reduce the complexity of using a multi‐cloud environment. In this paper, we show a cloud broker architecture for deploying virtualized servers across available clouds. This architecture uses a scheduling module to obtain an optimal placement of a virtual infrastructure while making it transparent for users. We focus our investigation on dynamic cloud scenarios in terms of variable resource prices, taking into account the virtual machine migration overhead issue. We evaluate the performance of several use cases applied in real scenarios, and we show the improvement potential of using brokering mechanisms in dynamic deployments compared with static ones. Copyright © 2012 John Wiley & Sons, Ltd.

This article gathers the foundational premises of the MEDIANET project as well as intermediate results obtained along its firsts two years. MEDIANET is a Spanish project founded by the Comunidad de Madrid government, which strives for a significant scientific advance in the future media Internet where important advances are necessary to allow end-users to perceive a good quality of experience. The network technologies objectives consist of the definition and validation of new proposals for the efficient transport of high bandwidth, real-time data flows in a decentralized way where the network provides mechanisms to seamlessly request and configure devices to increase the quality of experience perceived by end-users. Furthermore, new experiences with layer 2 networks and a cross-layer design will be tested with high bandwidth demanding media services. An important objective is to develop, evaluate in depth and implement on diverse platforms, a new low latency transparent bridge protocol based in on-demand path set up, suitable for campus and data center networks. The global result will be an integrated and independent advancement in future media Internet protocols, algorithms, switching architectures and standards.

In current distributed systems, such as Grids, Clouds, or P2P systems, the amount of information to handle influences the way the system is managed. In P2P systems containing large quantities of data, or in Grid systems containing a large number of (often heterogeneous) resources, information about data or resources must be spread through the system in an efficient way in order to allow them to be found. An information discovery technique based on data summarization, via clustering, is presented. These summaries can be used to classify information to provide users with greater insight about documents or computing resources compared to raw data. Also, meta-schedulers or brokers would benefit from the proposed technique due to the fact that they would have to deal with less data from resources, thus aiding to the scalability of the system. An evaluation of the approach is subsequently provided to identify the impact of choosing particular parameters to be used as part of the summary.

ISDEP (Integrator of Stochastic Differential Equations for Plasmas) is a Monte Carlo code that solves the plasma dynamics in a fusion device and perfectly scales on distributed computing platforms. Montera is a recent framework developed for achieving Grid efficient executions of Monte Carlo applications, as ISDEP is. In this work, the improvement of performing the calculations of ISDEP with Montera, which rise up to 34.9%, is shown as well as an analysis on the implications it could have, which aim to show to the fusion research community the benefits of using Montera.

Abstract Cluster‐based solutions are being widely adopted for implementing flexible, scalable, low‐cost and high‐performance web server platforms. One of the main difficulties to implement these platforms is the correct dimensioning of the cluster size, so as to satisfy variable and peak demand periods. In this context, virtualization is being adopted by many organizations as a solution not only to provide service elasticity, but also to consolidate server workloads, and improve server utilization rates. A virtualized web server can be dynamically adapted to the client demands by deploying new virtual nodes when the demand increases, and powering off and consolidating virtual nodes during periods of low demand. Furthermore, the resources from the in‐house infrastructure can be complemented with a cloud provider (cloud bursting), so that peak demand periods can be satisfied by deploying cluster nodes in the external cloud, on an on‐demand basis. In this paper, we analyze the scalability of hybrid virtual infrastructures for two different distributed web server cluster implementations: a simple web cluster serving static files and a multi‐tier web server platform running the CloudStone benchmark. Copyright © 2011 John Wiley & Sons, Ltd.

Cloud computing is gaining acceptance in many IT organizations, as an elastic, flexible, and variable-cost way to deploy their service platforms using outsourced resources. Unlike traditional utilities where a single provider scheme is a common practice, the ubiquitous access to cloud resources easily enables the simultaneous use of different clouds. In this paper, we explore this scenario to deploy a computing cluster on the top of a multicloud infrastructure, for solving loosely coupled Many-Task Computing (MTC) applications. In this way, the cluster nodes can be provisioned with resources from different clouds to improve the cost effectiveness of the deployment, or to implement high-availability strategies. We prove the viability of this kind of solutions by evaluating the scalability, performance, and cost of different configurations of a Sun Grid Engine cluster, deployed on a multicloud infrastructure spanning a local data center and three different cloud sites: Amazon EC2 Europe, Amazon EC2 US, and ElasticHosts. Although the testbed deployed in this work is limited to a reduced number of computing resources (due to hardware and budget limitations), we have complemented our analysis with a simulated infrastructure model, which includes a larger number of resources, and runs larger problem sizes. Data obtained by simulation show that performance and cost results can be extrapolated to large-scale problems and cluster infrastructures.

FAFNER is code developed by Lister which simulates by Monte Carlo methods the Neutral Beam Injection (NBI) technology, one of the most extended heating methods for fusion devices. To the date, FAFNER has been usually run at CIEMAT adapted to the TJ-II helical axis stellarator on shared memory Cray architecture machines. From this version, FAFNER has been ported to the Grid in the framework of the EGEE Project. At the same time, this work is the first step of a more ambitious target since the code can be now coupled to many others such as ion transport tools. In this paper, all these preliminary advances are described as well as the performance and portability gains obtained.

Abstract The main purpose of the present study is to analyze the suitability of performance‐based scheduling strategies for High Throughput Computing applications applied to Federated Grids consisting of several Grid infrastructures. The infrastructures forming aFederated Grid are classified as internal or external. This is because our scheduling policies map as much jobs as possible first to internal resources. In this way the mapping strategies save time and communication bandwidth by exploiting the location of the internal resources and the membership to their respective resource domains. At the same time, the policies avoid overloading external resources facilitating the cooperation between organizations to share resources. In this paper we consider three new algorithms for scheduling in situations where there are several Grid infrastructures of different types: enterprise, partner, or utility. Copyright © 2009 John Wiley & Sons, Ltd.

Using virtual machines as a resource provisioning mechanism offers multiple benefits, most recently exploited by "infrastructure-as-a-service" clouds, but also poses several scheduling challenges. More specifically, although we can use the suspend/resume/migrate capability of virtual machines to support advance reservation of resources efficiently, by using suspension/resumption as a preemption mechanism, this requires adequately modeling the time and resources consumed by these operations to ensure that preemptions are completed before the start of a reservation. In this work we present a model for predicting various runtime overheads involved in using virtual machines, allowing us to efficiently support advance reservations. We extend our lease management software, Haizea, to use this new model in its scheduling decisions, and we use Haizea with the OpenNebula virtual infrastructure manager so the scheduling decisions will be enacted in a Xen cluster. We present both physical and simulated experimental results showing the degree of accuracy of our model and the long-term effects of variables in our model on several workloads.

RESERVOIR project is developing an advanced system and service management approach that will serve as the infrastructure for cloud computing and communications and future Internet of services by creative coupling of service virtualization, grid computing, networking and service management techniques. This paper presents work in progress for the integration and management of such systems into a new generation of managed service infrastructure.

Grid Computing has proven to be a solution for big workflow execution, especially in Bioinformatics. However, Grid nature itself introduces overheads that make its use in many cases an unfeasible solution if considering wall-time. Different heuristics such as list scheduling, agglomeration and replication are available for optimizing workflow execution. In particular, the replication heuristics have been previously used in heterogeneous environments with good results. In this work, we analyze their use for workflow scheduling on Grid infrastructures. In particular, we study its applications to an intree workflow, generated by the distribution of the CD-HIT application. The experiments were conducted on a testbed made of resources from two different grids and results show a significant reduction of the workflow execution time.

A performance model, previously proposed to characterize the performance of grid infrastructures, is extended to evaluate federations of grids by aggregating their performance parameters. These parameters can then be used to develop scheduling policies based on them. The new model can be used to take scheduling decisions based on them and hence to aid in the development of scheduling policies. The model has been validated using the performance results obtained in the execution of a high throughput computing application on an enterprise grid composed of Globus toolkit Web service resources and a GridGateWay giving access to gLite resources from the EGEE infrastructure.

Bioinformatics is demanding more computational resources day after day. The problems proposed by this area are growing in such complexity that traditional computing systems are not able to face them. For solving complex problems which can be divided in tasks with dependencies, a workflow management system must be employed. In this paper, we introduce the use of the workflow management of the GridWay metascheduler for running a Bioinformatics application which implements a complex algorithm performing protein clustering in order to obtain non-redundant protein databases. The use of a general purpose meta-scheduling system will provide the application the fault-tolerance and advance scheduling capabilities needed to execute on a highly dynamic, heterogeneous and faulty environment. The execution results on a production Grid (the EGEE infrastructure) shows the dramatic impact of remote queue waiting times on the application performance; and the critical need of efficient re-scheduling capabilities.

Virtual machines are a promising technology to overcome some of the problems found in current grid infrastructures, like heterogeneity, performance partitioning or application isolation. In this work, we present straightforward deployment of virtual machines in globus grids. This solution is based on standard services and does not require additional middleware to be installed. Also, we assess the suitability of this deployment in the execution of a high throughput scientific application, the XMM-Newton scientific analysis system.

Detailed knowledge of macromolecular structure is essential for the understanding of how the cellular machines work. Rigid body fitting is the common way to interpret the information contained in a 3D electron microscope (3DEM) medium-low resolution map in terms of its available atomic structural components. This fitting process, termed multi-resolution docking, consists in localizing atomic resolution structures into the 3DEM map by means of an exhaustive search of all possible relative rotations and translations. This exhaustive search is a highly computing demanding process and several search queries are also typically needed to select good fitting structures. Here, we present a novel and efficient Grid approach for performing these docking searches. This approach has been designed over the Gridway meta-scheduler. Results showing the high efficiency achieved are discussed together with the corresponding analysis of the performance obtained. The experiments were conducted on a Grid testbed built up from resources inside EGEE (LCG version of the pre-WS Globus components), the European production-level grid infrastructure, and resources from a research testbed based on the Globus Toolkit 4 (Web services components)

Loop distribution is one of the most useful techniques to reduce the execution time of parallel applications. Traditionally, loop scheduling algorithms are implemented based on parallel programming paradigms such as MPI. This approximation presents three main disadvantages when applied in a grid environment, namely: (i) all resources must be simultaneously allocated to begin execution of the application; (ii) it is necessary to restart the whole application when a resource fails; (iii) it is not possible to add new resources to a currently running application. To overcome these limitations, we propose a new approach to implement loop distribution schemes in computational Grids. This approach is implemented using the distributed resource management application API (DRMAA) standard and the GridWay meta-scheduling framework. The efficiency of this approach to solve the Mandelbrot set problem is analyzed in a Globus-based research testbed

Abstract Grids offer a dramatic increase in the number of available processing and storing resources that can be delivered to applications. However, efficient job submission and management continue being far from accessible to ordinary scientists and engineers due to their dynamic and complex nature. This paper describes a new Globus based framework that allows an easier and more efficient execution of jobs in a ‘submit and forget’ fashion. The framework automatically performs the steps involved in job submission and also watches over its efficient execution. In order to obtain a reasonable degree of performance, job execution is adapted to dynamic resource conditions and application demands. Adaptation is achieved by supporting automatic application migration following performance degradation, ‘better’ resource discovery, requirement change, owner decision or remote resource failure. The framework is currently functional on any Grid testbed based on Globus because it does not require new system software to be installed in the resources. The paper also includes practical experiences of the behavior of our framework on the TRGP and UCM‐CAB testbeds. Copyright © 2004 John Wiley & Sons, Ltd.

The ability to have applications draw computing power from a global resource pool to achieve high performance has become a new challenge for distributed computing and Internet technologies. This challenge not only involves solving technical difficulties in the construction of Grid environments, it also involves resource sharing and performance concerns. This paper presents a resource performance manager that fits in the current Globus implementation of computational Grids, and shows how it can aid in the Grid's expansion, illustrated by some experiments carried out on the UCM-CAB testbed.

This paper presents an efficient parallel multigrid solver for speeding up the computation of a 3-D model that treats the flow of a viscous fluid over a solid obstacle. From a numerical point of view, the main interest of this simulation lies in exhibiting some basic difficulties that prevent optimal multigrid efficiencies from being achieved. As the computing platform we have used Coral, a Beowulf-class system based on a high performance GigaNet cLAN interconnect and Dual Pentium III nodes. The solver which has been devised taking both algorithmic and architectural issues into account, not only attains good convergence rates, but also achieves satisfactory efficiencies on the target cluster.

The aim of this paper is to study the effect of local memory hierarchy and communication network exploitation on message sending and the influence of this effect on the decomposition of regular applications. In particular, we have considered two different parallel computers, a Cray T3E-900 and an SGI Origin 2000. In both systems, the bandwidth reduction due to non-unit-stride memory access is quite significant and could be more important than the reduction due to contention in the network. These conclusions affect the choice of optimal decompositions for regular domains problems. Thus, although traditional 3D decompositions lead to lower inherent communication-to-computation ratios and could exploit more efficiently the interconnection network, lower dimensional decompositions are found to be more efficient due to the data decomposition effects on the spatial locality of the messages to be communicated. This increasing importance of local optimisations has also been shown using a well-known communication-computation overlapping technique which increases execution time, instead of reducing it as we could expect, due to poor cache memory exploitation.

The aim of this paper is the study of different methods for the solution of alternating-line problems, taking into account the evolution of architectural parameters on modern parallel computers, i.e. processors, memory hierarchy, and interconnection network performance. Three different kinds of solvers are studied: The Pipelined Gaussian Elimination scheme, the Matrix Transposition scheme, and the new one which is presented in this paper: the Mapping Transposition scheme, whose performance clearly betters, in many cases, that obtained by all the other methods, due to its better fitting to the characteristics of modern parallel computers. The experimental results have been obtained on a Cray T3E and on an SGI Origin 2000, up to 512 and 32 processors, respectively.

The large number of processing elements in current parallel systems necessitates the development of more comprehensive and realistic tools for the scalability analysis of algorithms on those architectures. This paper presents a simple analytical tool with which to study the scalability of parallel algorithm-architecture combinations. Our practical method studies separately execution time, efficiency, and memory usage in the accuracy-critical scaling model, where the problem size-input data set size-increases with the number of processors, which is the relevant one in many situations. The paper defines quantitative and qualitative measurements of the scalability and derives important relationships between execution time and efficiency. For example, results show that the best way to scale the system (to deteriorate as little as possible the properties of the system) is by maintaining constant execution time. These analytical results are verified with one candidate application for massive parallel computers: the full multigrid method. We study the scalability of a general d-dimensional full multigrid method on an r-dimensional mesh of processors. The analytical expressions are verified through experimental results obtained by implementing the full multigrid method on a Transputer-based machine and on the CRAY T3D.