The rapid growth of Artificial Intelligence and Machine Learning in scientific research has highlighted a gap between industry-standard machine learning operations (MLOps) tools and platforms and the unique requirements of modern and Open Science, particularly regarding the FAIR (Findable, Accessible, Interoperable, and Reusable) principles. This paper presents AI4EOSC, a federated, open-source platform designed to operationalize the full AI/ML life-cycle within the European Open Science Cloud (EOSC) ecosystem. Our methodology tackles the fragmentation of distributed research infrastructures by integrating a modular and distributed architecture comprising an AI development platform, a serverless AI-as-a-Service layer, and a federated orchestration model that is able to integrate heterogeneous computing and storage resources from distributed e-infrastructures. AI4EOSC also introduces a “FAIR-by-design” approach that enforces metadata standardization (via MLDCAT-AP) and W3C PROV-compliant provenance tracking through a platform-integrated CI/CD pipeline. The added value of AI4EOSC is demonstrated through the delivery of a diverse set of community installations, which show consistent and seamless deployment across heterogeneous cloud providers. These installations are validated by a set of scientific cases, showing how our work reduces the manual burden on researchers while ensuring high levels of reproducibility and interoperability and providing a unified environment for the development, training, and production of AI/ML models in the EOSC.

Data will be made available on request.

This paper proposes an automated framework for efficient application profiling and training of Machine Learning (ML) performance models, composed of two parts: OSCAR-P and aMLLibrary. OSCAR-P is an auto-profiling tool designed to automatically test serverless application workflows running on multiple hardware and node combinations in cloud and edge environments. OSCAR-P obtains relevant profiling information on the execution time of the individual application components. These data are later used by aMLLibrary to train ML-based performance models. This makes it possible to predict the performance of applications on unseen configurations. We test our framework on clusters with different architectures (x86 and arm64) and workloads, considering multi-component use-case applications. This extensive experimental campaign proves the efficiency of OSCAR-P and aMLLibrary, significantly reducing the time needed for the application profiling, data collection, and data processing. The preliminary results obtained on the ML performance models accuracy show a Mean Absolute Percentage Error lower than 30% in all the considered scenarios.

Serverless computing was a breakthrough in Cloud computing due to its high elasticity capabilities and fine-grained pay-per-use model offered by the main public Cloud providers. Meanwhile, open-source serverless platforms supporting the FaaS (Function as a Service) model allow users to take advantage of many of their benefits while operating on the on-premises platforms of organizations. This opens the possibility to deploy and exploit them on the different layers of the cloud-to-edge continuum, either on IoT (Internet of Things) devices located at the Edge (i.e. next to data acquisition devices), in on-premises clusters closer to the data sources (i.e. Fog computing) or directly on the Cloud. This paper presents two strategies to mitigate the overload that disparate data ingestion rates may cause in low-powered devices at the Edge or Fog layers. To this end, it is proposed to delegate and reschedule serverless jobs between the different layers of the cloud-to-edge continuum using an open-source platform for event-driven file processing. To demonstrate the performance of these strategies, a use case for fire detection is proposed that includes processing in the Fog via minified Kubernetes clusters located near the Edge, in the private Cloud via on-premises elastic clusters and, finally, in the public Cloud by using the AWS (Amazon Web Services) Lambda FaaS service. The results indicate that these strategies can mitigate overloads in use cases involving processing across the cloud-to-edge continuum by coordinating several layers of computing resources.

The advent of open-source serverless computing frameworks has introduced the ability to bring the Functions-as-a-Service (FaaS) paradigm for applications to be executed on-premises. In particular, data-driven scientific applications can benefit from these frameworks with the ability to trigger scalable computation in response to incoming workloads of files to be processed. This paper introduces an open-source framework to achieve on-premises serverless computing for event-driven data processing applications that features: i) the automated provisioning of an elastic Kubernetes cluster that can grow and shrink, in terms of the number of nodes, on multi-Clouds; ii) the automated deployment of a FaaS framework together with a data storage back-end that triggers events upon file uploads; iii) a service that provides a REST API to orchestrate the creation of such functions and iv) a graphical user interface that provides a unified entry point to interact with the aforementioned services. Together, this provides a framework to deploy a computing platform to create highly-parallel event-driven file-processing serverless applications that execute on customized runtime environments provided by Docker containers that run on an elastic Kubernetes cluster. The usefulness of this framework is exemplified by means of the execution of a data-driven workflow for optimised object detection on video. The workflow is tested under three different workloads which process ten, a hundred and a thousand functions. The results show that the presented architecture is able to process such workloads taking advantage of its elasticity to make a sensible usage of the resources.

After a sequence of creation and destruction of virtual machines (VMs) in an on-premises Cloud computing platform, the scheduling decisions to host the VMs are far from being optimal and the fragmentation of the physical resources may impede the platform to host some VMs despite the free available virtualization resources. This paper describes a Virtual Machine Consolidation Agent that addresses this problem by analyzing the distribution of the VMs in the virtualization platform to migrate some of them among hosts, in order to defragment the physical resources and to enhance the efficiency on their usage. The agent has been validated in a production platform, where it is capable of minimizing the number of servers needed to host the VMs. The algorithms achieve near-optimal values at a very reduced computational cost, thus making it suitable for production platforms.

There is an opportunity for Distributed Computing Infrastructures (DCIs) to embrace container-based virtualisation to support efficient execution of scientific applications without the performance penalty commonly introduced by Virtual Machines (VMs). However, containers (e.g. Docker) and VMs feature different image formats and disparate procedures for deployment and management, thus hindering the adoption of hybrid DCIs (HDCIs) comprised of those kind of resources. This paper describes a workflow based on open-source tools and standards to introduce coherent application delivery on HDCIs in which applications require to be deployed on both VMs and Docker containers. Leveraging and extending the TOSCA standard to describe application requirements, and adopting DevOps practices, resulted in the coherent creation of the artifacts required for the execution of the applications on different platforms. The paper features the adoption of this approach in the INDIGO-DataCloud project.

A case study on climate models intercomparison data analysis addressing several classes of multi-model experiments is being implemented in the context of the EU H2020 INDIGO-DataCloud project. Such experiments require the availability of large amount of data (multi-terabyte order) related to the output of several climate models simulations as well as the exploitation of scientific data management tools for large-scale data analytics. More specifically, the paper discusses in detail a use case on precipitation trend analysis in terms of requirements, architectural design solution, and infrastructural implementation. The experiment has been tested and validated on CMIP5 datasets, in the context of a large scale distributed testbed across EU and US involving three ESGF sites (LLNL, ORNL, and CMCC) and one central orchestrator site (PSNC).

This paper describes the research work in the context of the CLUVIEM project towards achieving migratable, self-managed virtual elastic clusters on hybrid Cloud infrastructures. These virtual clusters can span across on-premises and public Cloud infrastructures thus leveraging hybrid Cloud platforms. They are elastic since working nodes are automatically provisioned and relinquished to dynamically adapt the capacity of the virtual cluster (in terms of number of nodes) according to the current workload. They are self-managed since the elasticity rules are managed via the head node without requiring any external software entity for monitoring and deciding when to scale in and out. Finally, they are migratable since they consider both application migration, via application check pointing, and infrastructure migration, by cloning infrastructures across multi-Clouds. These features introduce unprecedented flexibility for cost-effective cluster-based computing with minimal impact for cluster users. The paper summarises the current state of developments and future roads to achieve this vision.

Summary This paper presents a software platform to dynamically deploy complex scientific virtual computing infrastructures, on top of Infrastructure as a Service Clouds. The platform orchestrates different services to provision the virtual computing resources. It dynamically installs the appropriate software to satisfy the requirements of a researcher, both on public and on‐premise Clouds. The platform provides a web interface to enable the users to easily manage the life cycle of virtual infrastructures. It enables users to define infrastructures, share them with other users, deploy and relinquish them, add or remove resources dynamically, create and share application recipes, and so on. The paper also describes three case studies to deploy complex infrastructures, namely, a Hadoop cluster, a single‐node to perform Next Generation Sequencing and a gateway for users to access the European Grid Infrastructure. This platform promotes a better use of on‐premise hardware resources of a research center by allocating the computing resources just‐in‐time to the specific life time of the virtual infrastructures as well as the deployment of the very same infrastructures on a public Cloud. Copyright © 2015 John Wiley & Sons, Ltd.

The increasing interest of online learning is unquestionable nowadays, with MOOCs being taken by thousands of students. However, for online learning to go mainstream it is necessary that professors perceive that the effort required to prepare and manage an online course is manageable. Today, a myriad of inexpensive tools and services can be used to produce and manage online courses with unprecedented ease and without distressing the professor. For that, this paper proposes an architecture based on Cloud services that simplifies the process of managing an online course, from delivering on-demand fully customized remote laboratories to communication automation for student engagement and feedback gathering. This approach has been applied to produce, distribute and manage an Online Course on Cloud Computing with Amazon Web Services. The paper describes the methodology, tools and results of this experience to point out that it is possible to deliver online courses with automatically provisioned labs, with minimal management overhead, while still providing a high quality learning experience to a worldwide audience.

This paper presents a general energy management system for HPC clusters and cloud infrastructures that powers off cluster nodes when they are not being used, and conversely powers them on when they are needed. This system can be integrated with different HPC cluster middleware, such as Batch-Queuing Systems or Cloud Management Systems, by using a set of connectors, and is also able to deal with different mechanisms for powering on and off the computing nodes (such as Wake-on-Lan, Power Device Units, Intelligent Platform Management Interface or other infrastructure-specific mechanisms). While some existing Batch-Queuing Systems provide energy saving mechanisms, other popular choices lack this feature. Cloud management middleware do not generally provide this feature out of the box, and incorporating it implies making modifications to the middleware. The advantage of our approach is that it can be integrated with different resource management middleware, without needing any modification of that middleware. The paper describes the successful integration of the system proposed with the popular Torque/PBS management system, and also with the OpenNebula open source cloud management tool. Two real use-cases are presented, involving two different HPC clusters. These use cases show significant energy/costs savings of 38% and 16%.

With the advent of cloud technologies the scientists have access to different cloud infrastructures in order to deploy all the virtual machines they need to perform the computations required in their research works. This paper describes a software architecture and a description language to simplify the creation of all the needed resources, and the elastic evolution of the computing infrastructure depending on the application requirements and some QoS features.

This paper presents an e-commerce tool that extends the conventional online store with a new section called room planner, a web application which is embedded in the virtual store. It allows the specification of the geometry of the room, placement of the objects and selection of the point of view. Then a realistic picture of the scene can be obtained. This functionality is very suitable for the furniture and ceramics sectors. In order to generate the images a parallel radiosity illumination algorithm has been implemented, which can be used in low-cost platforms such as a cluster of PCs, so that these technologies are affordable also for SMEs.