Search

Federated Learning (FL) is a distributed Machine Learning (ML) technique that can benefit from cloud environments while preserving data privacy. We propose Multi-FedLS, a framework that manages multi-cloud resources, reducing execution time and financial costs of Cross-Silo Federated Learning applications by using preemptible VMs, cheaper than on-demand ones but that can be revoked at any time. Our framework encloses four modules: Pre-Scheduling, Initial Mapping, Fault Tolerance, and Dynamic Scheduler. This paper extends our previous work \cite{brum2022sbac} by formally describing the Multi-FedLS resource manager framework and its modules. Experiments were conducted with three Cross-Silo FL applications on CloudLab and a proof-of-concept confirms that Multi-FedLS can be executed on a multi-cloud composed by AWS and GCP, two commercial cloud providers. Results show that the problem of executing Cross-Silo FL applications in multi-cloud environments with preemptible VMs can be efficiently resolved using a mathematical formulation, fault tolerance techniques, and a simple heuristic to choose a new VM in case of revocation., Comment: In review by Journal of Parallel and Distributed Computing

We propose to demonstrate a network slice placement optimization solution based on Deep Reinforcement Learning (DRL), referred to as Heuristically-controlled DRL, which uses a heuristic to control the DRL algorithm convergence. The solution is adapted to realistic networks with large scale and under non-stationary traffic conditions (namely, the network load). We demonstrate the applicability of the proposed solution and its higher and stable performance over a non-controlled DRL-based solution. Demonstration scenarios include full online learning with multiple volatile network slice placement request arrivals., Comment: arXiv admin note: substantial text overlap with arXiv:2010.08295

The evaluation of the impact of using Machine Learning in the management of softwarized networks is considered in multiple research works. Beyond that, we propose to evaluate the robustness of online learning for optimal network slice placement. A major assumption to this study is to consider that slice request arrivals are non-stationary. In this context, we simulate unpredictable network load variations and compare two Deep Reinforcement Learning (DRL) algorithms: a pure DRL-based algorithm and a heuristically controlled DRL as a hybrid DRL-heuristic algorithm, to assess the impact of these unpredictable changes of traffic load on the algorithms performance. We conduct extensive simulations of a large-scale operator infrastructure. The evaluation results show that the proposed hybrid DRL-heuristic approach is more robust and reliable in case of unpredictable network load changes than pure DRL as it reduces the performance degradation. These results are follow-ups for a series of recent research we have performed showing that the proposed hybrid DRL-heuristic approach is efficient and more adapted to real network scenarios than pure DRL., Comment: arXiv admin note: substantial text overlap with arXiv:2105.06741, arXiv:2108.02495

We consider online learning for optimal network slice placement under the assumption that slice requests arrive according to a non-stationary Poisson process. We propose a framework based on Deep Reinforcement Learning (DRL) combined with a heuristic to design algorithms. We specifically design two pure-DRL algorithms and two families of hybrid DRL-heuristic algorithms. To validate their performance, we perform extensive simulations in the context of a large-scale operator infrastructure. The evaluation results show that the proposed hybrid DRL-heuristic algorithms require three orders of magnitude of learning episodes less than pure-DRL to achieve convergence. This result indicates that the proposed hybrid DRL-heuristic approach is more reliable than pure-DRL in a real non-stationary network scenario., Comment: arXiv admin note: text overlap with arXiv:2105.06741

We present a hybrid ML-heuristic approach that we name "Heuristically Assisted Deep Reinforcement Learning (HA-DRL)" to solve the problem of Network Slice Placement Optimization. The proposed approach leverages recent works on Deep Reinforcement Learning (DRL) for slice placement and Virtual Network Embedding (VNE) and uses a heuristic function to optimize the exploration of the action space by giving priority to reliable actions indicated by an efficient heuristic algorithm. The evaluation results show that the proposed HA-DRL algorithm can accelerate the learning of an efficient slice placement policy improving slice acceptance ratio when compared with state-of-the-art approaches that are based only on reinforcement learning., Comment: arXiv admin note: substantial text overlap with arXiv:2105.06741

Network Slice placement with the problem of allocation of resources from a virtualized substrate network is an optimization problem which can be formulated as a multiobjective Integer Linear Programming (ILP) problem. However, to cope with the complexity of such a continuous task and seeking for optimality and automation, the use of Machine Learning (ML) techniques appear as a promising approach. We introduce a hybrid placement solution based on Deep Reinforcement Learning (DRL) and a dedicated optimization heuristic based on the Power of Two Choices principle. The DRL algorithm uses the so-called Asynchronous Advantage Actor Critic (A3C) algorithm for fast learning, and Graph Convolutional Networks (GCN) to automate feature extraction from the physical substrate network. The proposed Heuristically-Assisted DRL (HA-DRL) allows to accelerate the learning process and gain in resource usage when compared against other state-of-the-art approaches as the evaluation results evidence.

Leading Cloud providers offer several types of Virtual Machines (VMs) in diverse contract models, with different guarantees in terms of availability and reliability. Among them, the most popular contract models are the on-demand and the spot models. In the former, on-demand VMs are allocated for a fixed cost per time unit, and their availability is ensured during the whole execution. On the other hand, in the spot market, VMs are offered with a huge discount when compared to the on-demand VMs, but their availability fluctuates according to the cloud's current demand that can terminate or hibernate a spot VM at any time. Furthermore, in order to cope with workload variations, cloud providers have also introduced the concept of burstable VMs which are able to burst up their respective baseline CPU performance during a limited period of time with an up to 20% discount when compared to an equivalent non-burstable on-demand VMs. In the current work, we present the Burst Hibernation-Aware Dynamic Scheduler (Burst-HADS), a framework that schedules and executes tasks of Bag-of-Tasks applications with deadline constraints by exploiting spot and on-demand burstable VMs, aiming at minimizing both the monetary cost and the execution time. Based on ILS metaheuristics, Burst-HADS defines an initial scheduling map of tasks to VMs which can then be dynamically altered by migrating tasks of a hibernated spot VM or by performing work-stealing when VMs become idle. Performance results on Amazon EC2 cloud with different applications show that, when compared to a solution that uses only regular on-demand instances, Burst-HADS reduces the monetary cost of the execution and meet the application deadline even in scenarios with high spot hibernation rates. It also reduces the total execution time when compared to a solution that uses only spot and non-burstable on-demand instances.

We propose an online heuristic algorithm for the problem of network slice placement optimization. The solution is adapted to support placement on large scale networks and integrates Edge-specific and URLLC constraints. We rely on an approach called the Power of Two Choices to build the heuristic. The evaluation results show the good performance of the heuristic that solves the problem in few seconds under a large scale scenario. The heuristic also improves the acceptance ratio of network slice placement requests when compared against a deterministic online Integer Linear Programming (ILP) solution.

In this demo paper, we consider the network slice placement optimization problem and give some insights into a fast heuristic algorithm tailored to placement in large scale networks. We consider an online optimization scenario with multiple and volatile network slice request arrivals and we showcase the applicability of the proposed Edge-enabled network slice placement solution through a Proof-of-Concept illustrated by large scale networks scenarios.

Cloud platforms have emerged as a prominent environment to execute high performance computing (HPC) applications providing on-demand resources as well as scalability. They usually offer different classes of Virtual Machines (VMs) which ensure different guarantees in terms of availability and volatility, provisioning the same resource through multiple pricing models. For instance, in Amazon EC2 cloud, the user pays per hour for on-demand VMs while spot VMs are unused instances available for lower price. Despite the monetary advantages, a spot VM can be terminated, stopped, or hibernated by EC2 at any moment. Using both hibernation-prone spot VMs (for cost sake) and on-demand VMs, we propose in this paper a static scheduling for HPC applications which are composed by independent tasks (bag-of-task) with deadline constraints. However, if a spot VM hibernates and it does not resume within a time which guarantees the application's deadline, a temporal failure takes place. Our scheduling, thus, aims at minimizing monetary costs of bag-of-tasks applications in EC2 cloud, respecting its deadline and avoiding temporal failures. To this end, our algorithm statically creates two scheduling maps: (i) the first one contains, for each task, its starting time and on which VM (i.e., an available spot or on-demand VM with the current lowest price) the task should execute; (ii) the second one contains, for each task allocated on a VM spot in the first map, its starting time and on which on-demand VM it should be executed to meet the application deadline in order to avoid temporal failures. The latter will be used whenever the hibernation period of a spot VM exceeds a time limit. Performance results from simulation with task execution traces, configuration of Amazon EC2 VM classes, and VMs market history confirms the effectiveness of our scheduling and that it tolerates temporal failures.

In this work we present VCube-PS, a topic-based Publish/Subscribe system built on the top of a virtual hypercube-like topology. Membership information and published messages are broadcast to subscribers (members) of a topic group over dynamically built spanning trees rooted at the publisher. For a given topic, the delivery of published messages respects the causal order. VCube-PS was implemented on the PeerSim simulator, and experiments are reported including a comparison with the traditional Publish/Subscribe approach that employs a single rooted static spanning-tree for message distribution. Results confirm the efficiency of VCube-PS in terms of scalability, latency, number and size of messages., Comment: Improved text and performance evaluation. Added proof for the algorithms (Section 3.4)

Distributed storage systems such as Hadoop File System or Google File System (GFS) ensure data availability and durability using replication. This paper is focused on the analysis of the efficiency of replication mechanism that determines the location of the copies of a given file at some server. The variability of the loads of the nodes of the network is investigated for several policies. Three replication mechanisms are tested against simulations in the context of a real implementation of a such a system: Random, Least Loaded and Power of Choice. The simulations show that some of these policies may lead to quite unbalanced situations: if $\beta$ is the average number of copies per node it turns out that, at equilibrium, the load of the nodes may exhibit a high variability. It is shown in this paper that a simple variant of a power of choice type algorithm has a striking effect on the loads of the nodes: at equilibrium, the distribution of the load of a node has a bounded support, most of nodes have a load less than $2\beta$ which is an interesting property for the design of the storage space of these systems. Mathematical models are introduced and investigated to explain this interesting phenomenon. The analysis of these systems turns out to be quite complicated mainly because of the large dimensionality of the state spaces involved. Our study relies on probabilistic methods, mean-field analysis, to analyze the asymptotic behavior of an arbitrary node of the network when the total number of nodes gets large. An additional ingredient is the use of stochastic calculus with marked Poisson point processes to establish some of our results.

In its classical form, a consistent replicated service requires all replicas to witness the same evolution of the service state. Assuming a message-passing environment with a majority of correct processes, the necessary and sufficient information about failures for implementing a general state machine replication scheme ensuring consistency is captured by the {\Omega} failure detector. This paper shows that in such a message-passing environment, {\Omega} is also the weakest failure detector to implement an eventually consistent replicated service, where replicas are expected to agree on the evolution of the service state only after some (a priori unknown) time. In fact, we show that {\Omega} is the weakest to implement eventual consistency in any message-passing environment, i.e., under any assumption on when and where failures might occur. Ensuring (strong) consistency in any environment requires, in addition to {\Omega}, the quorum failure detector {\Sigma}. Our paper thus captures, for the first time, an exact computational difference be- tween building a replicated state machine that ensures consistency and one that only ensures eventual consistency.

This work proposes a new and flexible unreliable failure detector whose output is related to the trust level of a set of processes. By expressing the relevance of each process of the set by an impact factor value, our approach allows the tuning of the detector output, making possible a softer or stricter monitoring. The idea behind our proposal is that, according to an acceptable margin of failures and the impact factor assigned to processes, in some scenarios, the failure of some low impact processes may not change the user confidence in the set of processes, while the crash of a high impact factor process may seriously affect it. We outline the application scenarios and the proposed unreliable failure detector, giving a detailed account of the concept on which it is based., Comment: EDCC-2014, Fast-Abstracts, failure detector, impact factor, trust level

Unlike their model membrane counterparts, biological membranes are richly decorated with a heterogeneous assembly of membrane proteins. These proteins are so tightly packed that their excluded area interactions can alter the free energy landscape controlling the conformational transitions suffered by such proteins. For membrane channels, this effect can alter the critical membrane tension at which they undergo a transition from a closed to an open state, and therefore influence protein function in vivo. Despite their obvious importance, crowding phenomena in membranes are much less well studied than in the cytoplasm. Using statistical mechanics results for hard disk liquids, we show that crowding induces an entropic tension in the membrane, which influences transitions that alter the projected area and circumference of a membrane protein. As a specific case study in this effect, we consider the impact of crowding on the gating properties of bacterial mechanosensitive membrane channels, which are thought to confer osmoprotection when these cells are subjected to osmotic shock. We find that crowding can alter the gating energies by more than [Formula: see text] in physiological conditions, a substantial fraction of the total gating energies in some cases. Given the ubiquity of membrane crowding, the nonspecific nature of excluded volume interactions, and the fact that the function of many membrane proteins involve significant conformational changes, this specific case study highlights a general aspect in the function of membrane proteins.

Grid infrastructure is a large set of nodes geographically distributed and connected by a communication. In this context, fault tolerance is a necessity imposed by the distribution that poses a number of problems related to the heterogeneity of hardware, operating systems, networks, middleware, applications, the dynamic resource, the scalability, the lack of common memory, the lack of a common clock, the asynchronous communication between processes. To improve the robustness of supercomputing applications in the presence of failures, many techniques have been developed to provide resistance to these faults of the system. Fault tolerance is intended to allow the system to provide service as specified in spite of occurrences of faults. It appears as an indispensable element in distributed systems. To meet this need, several techniques have been proposed in the literature. We will study the protocols based on rollback recovery. These protocols are classified into two categories: coordinated checkpointing and rollback protocols and log-based independent checkpointing protocols or message logging protocols. However, the performance of a protocol depends on the characteristics of the system, network and applications running. Faced with the constraints of large-scale environments, many of algorithms of the literature showed inadequate. Given an application environment and a system, it is not easy to identify the recovery protocol that is most appropriate for a cluster or hierarchical environment, like grid computing. While some protocols have been used successfully in small scale, they are not suitable for use in large scale. Hence there is a need to implement these protocols in a hierarchical fashion to compare their performance in grid computing. In this paper, we propose hierarchical version of four well-known protocols. We have implemented and compare the performance of these protocols in clusters and grid computing using the Omnet++ simulator.

Grid infrastructure is a large set of nodes geographically distributed and connected by a communication. In this context, fault tolerance is a necessity imposed by the distribution that poses a number of problems related to the heterogeneity of hardware, operating systems, networks, middleware, applications, the dynamic resource, the scalability, the lack of common memory, the lack of a common clock, the asynchronous communication between processes. To improve the robustness of supercomputing applications in the presence of failures, many techniques have been developed to provide resistance to these faults of the system. Fault tolerance is intended to allow the system to provide service as specified in spite of occurrences of faults. It appears as an indispensable element in distributed systems. To meet this need, several techniques have been proposed in the literature. We will study the protocols based on rollback recovery. These protocols are classified into two categories: coordinated checkpointing and rollback protocols and log-based independent checkpointing protocols or message logging protocols. However, the performance of a protocol depends on the characteristics of the system, network and applications running. Faced with the constraints of large-scale environments, many of algorithms of the literature showed inadequate. Given an application environment and a system, it is not easy to identify the recovery protocol that is most appropriate for a cluster or hierarchical environment, like grid computing. While some protocols have been used successfully in small scale, they are not suitable for use in large scale. Hence there is a need to implement these protocols in a hierarchical fashion to compare their performance in grid computing. In this paper, we propose hierarchical version of four well-known protocols. We have implemented and compare the performance of these protocols in clusters and grid computing using the Omnet++ simulator.

The details of conformational changes undergone by transmembrane ion channels in response to stimuli, such as electric fields and membrane tension, remain controversial. We approach this problem by considering how the conformational changes impose deformations in the lipid bilayer. We focus on the role of bilayer deformations in the context of voltage-gated channels because we hypothesize that such deformations are relevant in this case as well as for channels that are explicitly mechanosensitive. As a result of protein conformational changes, we predict that the lipid bilayer suffers deformations with a characteristic free-energy scale of 10kBT. This free energy is comparable to the voltage-dependent part of the total gating energy, and we argue that these deformations could play an important role in the overall free-energy budget of gating. As a result, channel activity will depend upon mechanical membrane parameters such as tension and leaflet thickness. We further argue that the membrane deformation around any channel can be divided into three generic classes of deformation that exhibit different mechanosensitive properties. Finally, we provide the theoretical framework that relates conformational changes during gating to tension and leaflet thickness dependence in the critical gating voltage. This line of investigation suggests experiments that could discern the dominant deformation imposed upon the membrane as a result of channel gating, thus providing clues as to the channel deformation induced by the stimulus.

The details of conformational changes undergone by transmembrane ion channels in response to stimuli, such as electric fields and membrane tension, remain controversial. We approach this problem by considering how the conformational changes impose deformations in the lipid bilayer. We focus on the role of bilayer deformations in the context of voltage-gated channels because we hypothesize that such deformations are relevant in this case as well as for channels that are explicitly mechanosensitive. As a result of protein conformational changes, we predict that the lipid bilayer suffers deformations with a characteristic free-energy scale of 10kBT. This free energy is comparable to the voltage-dependent part of the total gating energy, and we argue that these deformations could play an important role in the overall free-energy budget of gating. As a result, channel activity will depend upon mechanical membrane parameters such as tension and leaflet thickness. We further argue that the membrane deformation around any channel can be divided into three generic classes of deformation that exhibit different mechanosensitive properties. Finally, we provide the theoretical framework that relates conformational changes during gating to tension and leaflet thickness dependence in the critical gating voltage. This line of investigation suggests experiments that could discern the dominant deformation imposed upon the membrane as a result of channel gating, thus providing clues as to the channel deformation induced by the stimulus.

We consider the problem of failure detection in dynamic networks such as MANETs. Unreliable failure detectors are classical mechanisms which provide information about process failures. However, most of current implementations consider that the network is fully connected and that the initial number of nodes of the system is known. This assumption is not applicable to dynamic environments. Furthermore, such implementations are usually timer-based while in dynamic networks there is no upper bound for communication delays since nodes can move. This paper presents an asynchronous implementation of a failure detector for unknown and mobile networks. Our approach does not rely on timers and neither the composition nor the number of nodes in the system are known. We prove that our algorithm can implement failure detectors of class <>S when behavioral properties and connectivity conditions are satisfied by the underlying system.

International audience, Cloud Computing provides a convenient means of remote on-demand and pay-per-use access to computing re- sources. However, its ad-hoc management of quality-of-service and SLA poses significant challenges to the performance, dependability and costs of online cloud services. The paper precisely addresses this issue and makes a threefold contribu- tion. First, it introduces a new cloud model, the SLAaaS (SLA aware Service) model. SLAaaS enables a systematic integration of QoS levels and SLA into the cloud. It is orthogonal to other cloud models such as SaaS or PaaS, and may apply to any of them. Second, the paper introduces CSLA, a novel language to describe QoS-oriented SLA associated with cloud services. Third, the paper presents a control-theoretic approach to provide performance, dependability and cost guarantees for online cloud services, with time-varying workloads. The proposed approach is validated through case studies and extensive experiments with online services hosted in clouds such as Amazon EC2. The case studies illustrate SLA guarantees for various services such as a MapReduce service, a cluster-based multi-tier e-commerce service, and a low-level locking service.

National audience, Le Cloud computing marque une nouvelle avancée vers l'infrastructure informatique dématérialisée. Le Cloud fournit des ressources informatiques, logicielles ou matérielles, accessible à distance en tant que service. L'adoption de ce modèle soulève un certain nombre de défis, notamment au sujet de la qua- lité de service (QoS) des services fournis. En effet, alors qu'on assiste à une multiplication de services similaires sur le Cloud (stockage, environnement de développement, bureautique, etc.), rien ne nous permet d'évaluer la qualité d'un service rendu. Cet article présente une solution pour intégrer la QoS et le contrat SLA (Service Level Agreement) comme éléments à part entière du Cloud. Les contributions de cet article sont les suivantes. Premièrement, nous proposons le modèle de Cloud SLAaaS (SLA aware as a Service) qui apporte une dimension supplémentaire au paradigme de Cloud Computing. Deuxiè- mement, nous présentons le langage Cloud Service Level Agreement (CSLA) qui permet la définition et l'établissement d'un contrat qui supporte l'instabilité de la QoS dans un environnement hautement dynamique comme le Cloud en gérant finement la violation. Enfin, nous développons deux stratégies dirigées par CSLA pour pouvoir contrôler dynamiquement et de manière autonome la taille et la forme du Cloud pour répondre aux contraintes de SLA : i) une stratégie de planification de capacité pour le redimensionnement sur mesure et ii) une stratégie de verrouillage d'exclusion mutuelle pour assurer un accès exclusif à des ressources partagées sur le Cloud.