In the age of globalization and international collaboration, collaborative data sharing is crucial for computer-aided research. There is a growing demand for spontaneous, short-lived, and fine-grained collaboration acts that could be satisfied by a decentralized platform for the management of entities (users, groups, datasets). Our research is based on a pre-existing concept of such a decentralized platform, called the decentralized entity persistence layer (DEPL). We describe a novel architecture extending DEPL, called DEPL-DA, which introduces an idea of decentralized authority, allowing entities to be managed collectively by multiple organizations, at the same time ensuring fault tolerance and improving availability. Our proposal utilizes a synergy between consensus and atomic commitment protocols (AC&C) providing decentralization of authority over entities, at the same time providing means to perform atomic modifications of entities spanning across multiple authoritative domains. We propose two approaches to the integration of AC&C: competitive and cooperative, and implement them based on several existing AC&C protocols. The evaluation shows the differences between approaches and protocols used, as well as the fact that both approaches are viable and offer latencies acceptable by operations related to entity management.

Summary Deep reinforcement learning has been recently a very active field of research. The policies generated with the use of this class of training algorithms are flexible and thus have many practical applications. In this article we present the results of our attempt to use the recent advancements in reinforcement learning to automate the management of resources in a compute cloud environment. We describe a new approach to self‐adaptation of autonomous management, which uses a digital clone of the managed infrastructure to continuously update the control policy. We present the architecture of our system and discuss the results of evaluation which includes autonomous management of a sample application deployed to Amazon Web Services cloud. We also provide the details of the training of the management policy using the Proximal Policy Optimization algorithm. Finally, we discuss the feasibility to extend the presented approach to further scenarios.

Abstract Many modern applications, both scientific and commercial, are deployed to cloud environments and often employ multiple types of resources. That allows them to efficiently allocate only the resources which are actually needed to achieve their goals. However, in many workloads the actual usage of the infrastructure varies over time, which results in over-provisioning and unnecessarily high costs. In such cases, automatic resource scaling can provide significant cost savings by provisioning only the amount of resources which are necessary to support the current workload. Unfortunately, due to the complex nature of distributed systems, automatic scaling remains a challenge. Reinforcement learning domain has been recently a very active field of research. Thanks to combining it with Deep Learning, many newly designed algorithms improve the state of the art in many complex domains. In this paper we present the results of our attempt to use the recent advancements in Reinforcement Learning to optimize the cost of running a compute-intensive evolutionary process by automating the scaling of heterogeneous resources in a compute cloud environment. We describe the architecture of our system and present evaluation results. The experiments include autonomous management of a sample workload and a comparison of its performance to the traditional automatic threshold-based management approach. We also provide the details of training of the management policy using the proximal policy optimization algorithm. Finally, we discuss the feasibility to extend the presented approach to further scenarios.

In order to realize collaboration on a global scale, academic research requires often large quantities of data to be shared between geographically dispersed organizations.The requirement to protect and govern data in a network of loosely coupled, autonomous institutions is an incentive for decentralized solutions, where the participants are in full control of their data without trusting a third-party provider to store and process the data.In order to increase data availability and fault tolerance in decentralized collaborative systems, we propose a layer, which is based on replication and decentralized authority over the data.The solution consists of an idea of peer-sets, which are groups of peers implementing collective data management, a consensus protocol which synchronizes a distributed ledger between peers, and an atomic commitment protocol used to implement optional two-way references between documents.This architecture may be utilized in various decentralized collaborative data-sharing systems, such as Onedata.

The observation of current trends in data access, especially in the field of scientific computations, shows that global data access that crosses federation boundaries is highly desirable. However, administrative constraints require that data centers remain autonomous, which effectively eliminates the possibility of cooperation. To overcome this, we plan to establish an open network of cooperating data providers. In this paper, we address the issue of data access control for such network. Our proposition is to use a synergy of hybrid peer-to-peer architecture, decentralized identity and access management, metadata synchronization protocol and trust driven authorization flow. The proposed solution is discussed using real-life use-cases concerning cross-federation data access.

Summary Sensitivity analysis is widely used in numerical simulations applied in industry. The robustness of such applications is crucial, which means they have to be fast and precise at the same time. However, the conventional approach to sensitivity analysis assumes realization of multiple execution of computationally intensive simulations to discover input/output dependencies. In this paper, we present a novel computational approach for performing large‐scale sensitivity analysis integrated with an extended platform for parameter studies – Scalarm – to make use of modern e‐infrastructures for distribution and parallelization purposes, profitable for complicated industrial problems. Copyright © 2016 John Wiley & Sons, Ltd.

This paper describes an approach to conduct large-scale parameter studies, where each data point in the study requires the execution of a whole scientific workflow. We show how a parameter studies system can be integrated with a workflow management system to seamlessly execute a large number of workflows, each with different input parameter values using large-scale computing infrastructure. The work is motivated by a need to collect performance-related data to conduct a sensitivity analysis in the context of relation between workflow input parameters and the performance of tasks in the workflow developed for the Spallation Neutron Source facility at the Oak Ridge National Laboratory.

Using multiple Clouds as a single environment to conduct simulation-based virtual experiments at a large-scale is a challenging problem. This paper describes how this can be achieved with the Scalarm platform in the context of data farming. In particular, a use case with a private Cloud combined with public, commercial Clouds is studied. We discuss the current architecture and implementation of Scalarm in terms of supporting different infrastructures, and propose how it can be extended in order to attain a unification of different Clouds usage. We discuss different aspects of the Cloud usage unification including: scheduling virtual machines, authentication, and virtual machine state monitoring. An experimental evaluation of the presented solution is conducted with a genetic algorithm solving the well-known Travel Salesman Problem. The evaluation uses three different resource configurations: using only public Cloud, using only private Cloud, and using both public and private Clouds.

Big science needs synergistic approach from theory, experiment and computations. Significance of the computational science is still on its way up due to the importance of results concerning features of extreme time and space scales as well as of multiscale phenomena being a subject of current research. These needs have been recognized internationally and one of the activities undertaken in the last decade is the European Grid Infrastructure, consolidating national efforts on computational infrastructures developed under National Grid Initiatives (NGI). In the paper we present the way of building Polish NGI established for Polish scientists, allowing them to conduct interdisciplinary research on a national scale, and giving them transparent access to international grid resources via affiliated international grid infrastructures. The approach applied in setting up this e-infrastructure ensured integration with other similar platforms around the world, recently with special focus on domain specific solutions.

The main intention of this paper is to introduce a new role-based approach to self-healing monitoring. Together with that we are going to provide an overview of the existing approaches to the monitoring of distributed systems using self healing features. Starting with a discussion of autonomous monitoring systems - the ones that do not need any user interaction to make a decision, we will come to self-healing systems. These systems should be able to automatically resolve the problems that occur in a system under monitoring. The paper provides insight into various aspects of self-healing monitoring systems at the software and hardware level. Later on we identify requirements for a new system. A detailed description of the new system, we call AgeMon, is covered at the end of the article. The system is based on the roles that can be fulfilled by the agents. The self-healing in the monitoring and monitored system can be achieved by the cooperation of agents performing different roles (like monitoring agent, rule agent, database agent). The article addresses the roles in the system together with the implementation background.

Growing role of the cooperation between companies around the world causes the rise of meaning of the SLA agreements which are negotiated among companies. These agreements usually require a reliable and robust monitoring process, since their fulfilment is crucial for the coherence of the cooperation. In this paper we present the application of the ESB integration architecture in the specialized distributed monitoring system-the FiVO SLA monitoring system-which is used to monitor the resources and services shared among the cooperating companies. Different deployment models of this system are proposed and the results from their performance tests are presented.

Summary form only given. As number and diversity of data, information and resources increase importance of usage of knowledge for their accessing and reuse grows significantly. One of the most useful approaches is based on semantic description of elementary data and pieces of information as well as on relations between them, therefore resulting in ontological approach. Semantics is starting to be used in wider scope than ever, especially for accessing distributed resources. This field is well represented by the Semantic Web, where WWW content can be expressed also in a format that can be read and used by automated tools, thus permitting people and machines to find, share and integrate information more easily. Semantic description of services and user requests can also be used in (semi-)automatic construction of distributed/Grid applications, often in a form of workflows. In order to effectively manage ontologies on the level higher than the local one, e.g., using Virtual Organization approach some sort of knowledge base must be provided.

In this paper we present an implementation and resultsof nonequilibrium particle dynamics simulations on CellBroadband Engine (CBE). As the particle interaction modelDissipative Particle Dynamics (DPD) method was chosen.One of the fundamental problems in computer simulationof models such as DPD is the determination of particle interaction.In order to compute all forces acting on a singleparticle a set of neighbours in a given proximity has tobe found.We propose an efficient method of task distributionbetween processor cores as well as fastcommunicationschemes. Our parallel DPD code is run on a single Playstation3 game console with Linux operating system, using upto over 700 · 105particles. We compare timings with thoseof traditional CPUs. Sample results from DPD simulationsconcern Rayleigh-Taylor instability.

Recent developments in the field of Grid computing reveal the strong need of managing metadata on much higher level than it used to be done previously. Standard solutions like MDS provide relatively efficient means of managing Grid resources on a low level basis, usually available for Grid experts and simple search tools. However in order to manage resources among multiple Virtual Organizations composed of highly heterogeneous entities, joining and leaving the Grid in a dynamic manner, a more flexible way of describing and managing their resources is required. In this paper, a solution to this problem based on ontologies and distributed knowledge base is presented. The knowledge base distributed architecture is discussed with the flexibility and fault-tolerance outlined. The performance evaluation for typical use cases shows, that ontology separation and various configurations of GOM make it useful for managing knowledge in multiple domains.

Benefits of using ontologies for automatic composition of workflow application, become evident when considering knowledge evolution influencing possibility of new workflows composition. The evolution of knowledge in the case of ontology descriptionmeans both evolution of the instance registries as well as evolution of schemas to which the instance descriptions conform. An approach to semi-automated knowledge evolution referred to ontology schemas applied to the process of new Grid service registration is introduced in the paper. The presented study makes extensive use of ontology alignment methods.

In this paper three algorithms for Molecular Dynamics simulation suitable for vector computers are reviewed and timings presented. The algorithms were tested on the ETA 10-P and IBM 3090/150E/VF supercomputers and compared against the INTEL 80386/80387 and MC 68020/68881 based microcomputers.