Search

Deep learning (DL) algorithms have already proved their effectiveness on a wide variety of application domains, including speech recognition, natural language processing, and image classification. To foster their pervasive adoption in applications where low latency, privacy issues and data bandwidth are paramount, the current trend is to perform inference tasks at the edge. This requires deployment of DL algorithms on low-energy and resource-constrained computing nodes, often heterogenous and parallel, that are usually more complex to program and to manage without adequate support and experience. In this paper, we present ALOHA, an integrated tool flow that tries to facilitate the design of DL applications and their porting on embedded heterogenous architectures. The proposed tool flow aims at automating different design steps and reducing development costs. ALOHA considers hardware-related variables and security, power efficiency, and adaptivity aspects during the whole development process, from pre-training hyperparameter optimization and algorithm configuration to deployment.

In process algebras such as ACP (Algebra of Communicating Processes), parallel processes are considered to be interleaved in an arbitrary way. In the case of multi-threading as found in contemporary programming languages, parallel processes are actually interleaved according to some interleaving strategy. An interleaving strategy is what is called a process-scheduling policy in the field of operating systems. In many systems, for instance hardware/software systems, we have to do with both parallel processes that may best be considered to be interleaved in an arbitrary way and parallel processes that may best be considered to be interleaved according to some interleaving strategy. Therefore, we extend ACP in this paper with the latter form of interleaving. The established properties of the extension concerned include an elimination property, a conservative extension property, and a unique expansion property., Comment: 19 pages, this version is a revision of the published version

We address the problem of detecting highly raised crowd density in situations such as indoor dance events.We propose a new method for estimating crowd density by anonymous, non-participatory, indoor Wi-Fi localization of smart phones. Using a probabilistic model inspired by statistical mechanics, and relying only on big data analytics, we tackle three challenges: (1) the ambiguity of Wi-Fi based indoor positioning, which appears regardless of whether the latter is performed with machine learning or with optimization, (2) the MAC address randomization when a device is not connected, and (3) the volatility of packet interarrival times. The main result is that our estimation becomes more -- rather than less -- accurate when the crowd size increases. This property is crucial for detection of dangerous crowd density.

In the embedded computer system domain, MPSoC systems have become increasingly popular due to the ever-increasing performance demands of modern embedded applications. The number of processing elements in these MPSoCs also steadily increases. Whereas current MPSoCs still contain a limited number of processing elements, future MPSoCs will feature tens up to hundreds of (heterogeneous) processing elements that are all integrated on a single chip. On these future large-scale MPSoC systems, the mapping of applications onto the hardware resources plays an important role to fully explore the parallelism of applications. In this article, a hierarchical run-time adaptive resource allocation framework which uses an intelligent task remapping approach is proposed to improve the system performance for large-scale MPSoCs.

In this paper we present a methodology for gaining a better understanding of the contribution behavior, interests and expertise of communities of Wikipedia users. Starting from a list of core articles and their main editors, we identify which other articles (outside of the initial list) they contributed to ‘significantly’. The ordering is based on (empirical) Bayesian estimates of the contribution probabilities for each of the articles. By constructing a co-contribution network, we can identify the general themes the community expresses exceptional interest (or disinterest) in. In order to show what type of insights one might gain from employing the proposed method, we use the editors that contributed to the articles on designer drugs as a case study. We find that the users in this community contribute significantly to articles on pharmaceuticals, popular party drugs, chemistry, mental illnesses, diseases, medicine and cell biology.

The abundance and heterogeneity of IT resources available, together with the ability to dynamically scale applications poses significant usability issues to users. Without understanding the performance profile of available resources users are unable to efficiently scale their applications in order to meet performance objectives. High quality media collaborations, like CineGrid, are one example of such diverse environments where users can leverage dynamic infrastructures to move and process large amounts of data. This paper describes our user-centric approach to executing high quality media processing workloads over dynamic infrastructures. Our main contribution is the CGtoolkit ?environment, an integrated system which aids users cope with the infrastructure complexity and large data sets specific to the digital cinema domain. We describe the general challenges user face in heterogeneous it infrastructure environments and identify their instance in the case of CineGrid.We characterize the most frequent workloads in the CineGrid collaboration including their infrastructure resource requirements.We present the design and implementation of a system which simplifies workload resource definition by leveraging semantic web description languages for infrastructure.We provide Pareto-optimal resource configurations for workload execution.

Increases in graph size and analytics complexity have brought graph processing at the forefront of HPC. However, the HPC shift towards manycore accelerators (e.g., GPUs) is not favourable: traditional graph processing is hardly suitable for regular parallelism. Previous work has focused on parallel algorithms for specific graph operations, often using assumptions about the structure of the input graph. However, there has been very little systematic investigation of how strongly a graph’s structure impacts the efficiency of graph operations.With this article we make propose a step to quantify this impact, focusing on a typical operation: neighbour iteration. We design and implement four strategies for neighbour iteration and introduce a simple model to reason about the expected impact of a graph’s structure on the performance of each strategy. We then use the PageRank algorithm to validate our model. We show that performance is significantly affected by the ability to effectively load-balance the work performed by these strategies across the GPU’s cores.

The multiclass queueing network (McQN) arises as a natural multiclass extension of the traditional (single-class) Jackson network. In a single-class network, subcriticality (i.e., subunitary nominal workload at every station) entails stability, but this is no longer sufficient when jobs/customers of different classes (i.e., with different service requirements and/or routing scheme) visit the same server; therefore, analytical conditions for stability of McQNs are lacking, in general. In this note, we design a numerical (simulation-based) method for determining the stability region of a McQN, in terms of arrival rate(s). Our method exploits certain (stochastic) monotonicity properties enjoyed by the associated Markovian queue-configuration process. Stochastic monotonicity is a quite common feature of queueing models and can be easily established in the single-class framework (Jackson networks); recently, also for a wide class of McQNs, including first-come-first-serve networks, monotonicity properties have been established. Here, we provide a minimal set of conditions, under which the method performs correctly. Eventually, we illustrate the use of our numerical method by presenting a set of numerical experiments, covering both single- and multiclass networks.

Part 12: Service Specification and Composition; International audience; The upward integration endeavor is making informatics systems (I-systems) increasingly complex. The modeling techniques, methodologies, development strategies, deployment and execution environment, maintenance and evolution, and governance, to mention just a few aspects are making the resulted (un)integrated informatics technology system a vendor lock-in landscape. The relation between informatics science and engineering and the organization’s business or control processes automation, or services provisioning and adaptation, has demonstrated to be difficult to converge to a common understanding of clear computational responsibility borders. Existing approaches and standards fail to be complete with respect to establishing a landscape of informatics technology under vendor agnostic model (lock-in free). In this context, this paper extends previous research by proposing an organization´s level modularity framework aiming at formally identifying an agnostic, and open informatics system of systems (ISoS). A definition of its components is provided, and a validation case study is discussed.

Secure Autonomous Response NETworks (SARNET) is a framework for auto­mated response against attacks on computer network infrastructures. The framework addresses several cyber-security problems at three crucial levels: strategic, tactical and operational.

Supporting exploratory search tasks with the help of structured data is an effective way to go beyond keyword search, as it provides an overview of the data, enables users to zoom in on their intent, and provides assistance during their navigation trails. However, finding a good starting point for a search episode in the given structure can still pose a considerable challenge, as users tend to be unfamiliar with exact, complex hierarchical structure. Thus, providing lookahead clues can be of great help and allow users to make better decisions on their search trajectory.In this paper, we investigate the behaviour of users when a recommendation engine is employed along with the browsing tool in an exploratory search system. We make use of an exploratory search system that facilitates browsing by mapping the data on a hierarchical structure. We designed and developed a path recommendation engine as a feature for this system, which given a text query, ranks different browsing paths in the hierarchy based on their likelihood of covering relevant documents. We conduct a user study comparing the baseline system with the featured system.Our main findings are as follows: We observe that, using the baseline system the users tend to explore the data in a breadth-first-like approach by visiting different data points at the same level of abstraction to choose one of them to expand and go deeper. Conversely, with browsing path recommendation (BPR) as a feature, the users tend to drive their search in a more depth-first-like approach by quickly going deep into the data hierarchy. While the users still incline to explore different parts of the search space by using BPR, they are able to restrain or augment their search focus more quickly and access smaller but more promising regions of the data. Therefore, they can complete their tasks with less time and effort

High performance graph processing poses significant challenges for both algorithm and platform designers due to the large performance variability it exhibits: performance depends on the algorithm, the dataset, and the (hardware/software) platform. Traditionally, performance variability is tackled by extensive benchmarking, modeling, and, eventually, better or smarter algorithms. Our own research into the impact of datasets on the performance of graph algorithms has convinced us that such extensive benchmarking is very difficult for graph processing simply because we lack input data: the public datasets and graph generation tools currently available are insufficient for a systematic investigation of the impact of different graph properties. In this work we propose to alleviate this problem by using evolutionary computing as a method to generate graphs. Our goal is allow users to request graphs with a given set of properties (e.g., number of vertices, number of edges, degree distribution), and enable the generator to evolve graphs until they satisfy the request. Such a fine-grain, flexible exploration of graphs and their properties will finally enable a statistical take on performance analysis and modeling for graph processing. To this end, our work-in-progress paper presents the design of our generator, discusses the challenges and trade-offs to be encountered, and reveals our preliminary results.

Software abstractions are crucial to effectively program heterogeneous Multi-Processor Systems on Chip (MPSoCs). Prime examples of such abstractions are Kahn Process Networks (KPNs) and execution traces. When modeling computation as a KPN, one of the key challenges is to obtain a good mapping, i.e., an assignment of logical computation and communication to physical resources. In this paper we compare two system-level frameworks for solving the mapping problem: Sesame and MAPS. These frameworks, while superficially similar, embody different approaches. Sesame, motivated by modeling and design-space exploration, uses evolutionary algorithms for mapping. MAPS, being a compiler framework, uses simple and fast heuristics instead. In this work we highlight the value of common abstractions, such as KPNs and traces, as a vehicle to enable comparisons between large independent frameworks. These types of comparisons are fundamental for advancing research in the area. At the same time, we illustrate how the lack of formalized models at the hardware level are an obstacle to achieving fair comparisons. Additionally, using a set of applications from the embedded systems domain, we observe that genetic algorithms tend to outperform heuristics by a factor between 1× and 5×, with notable exceptions. This performance comes at the cost of a longer computation time, between 0 and 2 orders of magnitude in our experiments.

This paper presents results of the ongoing development of the Cloud Services Delivery Infrastructure (CSDI) that provides a basis for infrastructure centric cloud services provisioning, operation and management in multi-cloud multi-provider environment defined as a Zero Touch Provisioning, Operation and Management (ZTP/ZTPOM) model. The presented work refers to use cases from data intensive research that require high performance computation resources and large storage volumes that are typically distributed between datacenters often involving multiple cloud providers. Automation for large scale scientific (and industrial) applications should include provisioning of both inter-cloud network infrastructure and intra-cloud application resources. It should provide support for the complete application operation workflow together with the possible application infrastructure and resources changes that can occur during the application lifecycle. The authors investigate existing technologies for automation of the service provisioning and management processes aiming to cross-pollinate best practices from currently disconnected domains such as cloud based applications provisioning and multi-domain high-performance network provisioning. The paper refers to the previous and legacy research by authors, the Open Cloud eXchange (OCX), that has been proposed to address the last mile problem in cloud services delivery to campuses over trans-national backbone networks such as GEANT. OCX will serve as an integral component of the prospective ZTP infrastructure over the GEANT network. Another important component, the Marketplace, is defined for providing cloud services and applications discovery (in generally intercloud environment) and may also support additional services such as services composition and trust brokering for establishing customer-provider federations.

New generations of many-core hardware become available frequently and are typically attractive extensions for data-centers because of power-consumption and performance benefits. As a result, supercomputers and clusters are becoming heterogeneous and start to contain a variety of many-core devices. Obtaining performance from a homogeneous cluster-computer is already challenging, but achieving it from a heterogeneous cluster is even more demanding. Related work primarily focuses on homogeneous many-core clusters. In this paper we present Cashmere, a programming system for heterogeneous many-core clusters. Cashmere is a tight integration of two existing systems: Satin is a programming system that provides a divide-and-conquer programming model with automatic load-balancing and latency-hiding, while Many-Core Levels is a programming system that provides a powerful methodology to optimize computational kernels for varying types of many-core hardware. We evaluate our system with several classes of applications and show that Cashmere achieves high performance and good scalability. The efficiency of heterogeneous executions is comparable to the homogeneous runs and is >90% in three out of four applications.

In the contemporary cloud computing environment, the various cloud providers adopt different architectures for their systems and usually these are not compatible. The scalable exchange of information about these heterogeneous resources is often not addressed; still this represents an interesting challenge, since different Cloud systems and vendors have different ways to describe and invoke their services, to specify and communicate requirements. Hence a way to provide a common access to Cloud services and to discover and use required services in Cloud federations is highly desirable. The IEEE P2302 Intercloud approach is to define a scalable architecture and to adopt the OMN (Open-Multinet) ontology for supporting its operation. Our work specifically describes the inner workings of the Intercloud system in terms of the Intercloud Federation Application Programmers Interface (API), which are in turn based on the semantic definition of resources, Service Level Agreements (SLAs), and Bearer Network Provisioning Metadata.

CloudComputing has several provisionmodels, e.g. Infrastructure as a service (IaaS). However, cloud users (tenants) have limited or no control over the underlying network resources. Network as a Service (NaaS) is emerging as a novelmodel to fill this gap.However,NaaS requires an approach capable of modeling the underlying network resources capabilities in abstracted and vendor-independent form. In this paper we elaborate on SemNaaS, a Semantic Web based approach for developing and supporting operations of NaaS systems. SemNaaS can workwith any NaaS provider. We integrated it with the existing OpenNaaS framework. We propose Network Markup Language (NML) as the ontology for describing networking infrastructures.Based on that ontology, we develop a network modeling system and integrate with OpenNaaS. Furthermore, we demonstrate the capabilities that Semantic Web can add to the NaaS paradigm by applying SemNaaS operations to a specific NaaS use case.

The Internet remains an unfinished work. There are several approaches to enhancing it that have been experimentally validated within federated testbed environments. To best gain scientific knowledge from these studies, reproducibility and automation are needed in all areas of the experiment life cycle. Within the GENI and FIRE context, several architectures and protocols have been developed for this purpose. However, a major open research issue remains, namely the description and discovery of the heterogeneous resources involved. To remedy this, we propose a semantic information model that can be used to allow declarative interoperability, build dependency graphs, validate requests, infer knowledge and conduct complex queries. The requirements for such an information model have been extracted from current international Future Internet research projects and the practicality of the model is being evaluated through initial implementations. The main outcome of this work is the definition of the Open-Multinet Upper Ontology and related sub-ontologies, which can be used to describe and manage federated infrastructures and their resources.

Applying suitable application models is essential to achieve efficient execution of the applications, effective development process and assure application portability and reusability. We observe that execution environments and the cloud environment in particular, lack of tools that would make it more available from a programmer perspective. Complexity and variety of libraries and authentication methods do not correspond with simplistic functionality of those services. We approach this issue at several fields - workflow systems, domain specific languages and cloud computing. Our work focus on two points, optimizing existing application models by supplementing them with additional tools and providing new features, and develop new models that are based on our research on applications in the cloud environment.

Estimation of the execution time is an important part of the workflow scheduling problem. The aim of this paper is to highlight common problems in estimating the workflow execution time and propose a solution that takes into account the complexity and the randomness of the workflow components and their runtime. The solution proposed in this paper addresses the problems at different levels from task to workflow, including the error measurement and the theory behind the estimation algorithm. The proposed estimation algorithm can be integrated easily into a wide class of schedulers as a separate module. We use a dual stochastic representation, characteristic / distribution functions, in order to combine tasks' estimates into the overall workflow makespan. Additionally, we propose the workflow reductions - the operations on a workflow graph that do not decrease the accuracy of the estimates, but simplify the graph structure, hence increasing the performance of the algorithm.

Storing data in the cloud is becoming a common trend, for both end-customers and data center operators. We propose a method for deciding where to host data storage tasks under the constraint of minimal greenhouse gas emission. The decision on whether to store data locally or store it remotely at a cleaner data center relies on the models for the local and remote data centers and the network connecting them. We conclude that the transport network that connects a local node and a 'cleaner' remote data center plays a significant role in the decision of where to store data, and that the frequency of access of the data is an important and related factor. © 2014 IEEE.

Computational and storage tasks can nowadays be offloaded among data centers, in order to optimize costs and or performance. We set out to investigate what are the environmental effects, namely the total CO2 emission, of such offloading. We built models for the various components present in these offloading scenarios, and we incorporated these models into an online calculator available to users. Our online calculator determines the total CO2 emitted by performing computational or storage tasks at a local or remote data center, provided that the energy sources of the data centers and the network connecting them are known. The demonstration we propose will allow conference visitors to use our website/calculator to get a sense for the impact of offloading their storage and computational tasks.

Scientific applications are frequently modeled as a workflow that is executed under the control of a workflow management system. One crucial requirement during the execution is the validation of the generated results and the traceability of the experiment execution path. The automated tracking and storage of provenance information during workflow execution could satisfy this requirement. To collect provenance data using the Grid-enabled scientific workflow management system WS-VLAM, experimentation was made with two different implementations of the provenance concepts. The first one adopts the Open Provenance Model (OPM) as basis to represent, store, and share scientific experiments metadata using the Provenance Layer Infrastructure for e-Science Resources (PLIER). The second one is the history-tracing XML (HisT) which was developed for e-Business provenance. HisT provides a specific model to store provenance data within layered XML documents, whereby each layer is related to one individual workflow task. This paper explores these two provenance models by using an example workflow application and describes how they are integrated into WS-VLAM including implementation details of the provenance architecture. It finally gives a comparison of the two different approaches with a special regard to provenance for human actors.