Search

Resource provisioning plays a pivotal role in determining the right amount of infrastructure resource to run applications and target the global decarbonization goal. A significant portion of production clusters is now dedicated to long-running applications (LRAs), which are typically in the form of microservices and executed in the order of hours or even months. It is therefore practically important to plan ahead the placement of LRAs in a shared cluster so that the number of compute nodes required by them can be minimized to reduce carbon footprint and lower operational costs. Existing works on LRA scheduling are often application-agnostic, without particularly addressing the constraining requirements imposed by LRAs, such as co-location affinity constraints and time-varying resource requirements. In this paper, we present an affinity-aware resource provisioning approach for deploying large-scale LRAs in a shared cluster subject to multiple constraints, with the objective of minimizing the number of compute nodes in use. We investigate a broad range of solution algorithms which fall into three main categories: Application-Centric, Node-Centric, and Multi-Node approaches, and tune them for typical large-scale real-world scenarios. Experimental studies driven by the Alibaba Tianchi dataset show that our algorithms can achieve competitive scheduling effectiveness and running time, as compared with the heuristics used by the latest work including Medea and LraSched., 22 pages, 7 figures

Federated learning allows multiple clients to collaborate to train high-performance deep learning models while keeping the training data locally. However, when the local data of all clients are not independent and identically distributed (i.e., non-IID), it is challenging to implement this form of efficient collaborative learning. Although significant efforts have been dedicated to addressing this challenge, the effect on the image classification task is still not satisfactory. In this paper, we propose FedProc: prototypical contrastive federated learning, which is a simple and effective federated learning framework. The key idea is to utilize the prototypes as global knowledge to correct the local training of each client. We design a local network architecture and a global prototypical contrastive loss to regulate the training of local models, which makes local objectives consistent with the global optima. Eventually, the converged global model obtains a good performance on non-IID data. Experimental results show that, compared to state-of-the-art federated learning methods, FedProc improves the accuracy by $1.6\%\sim7.9\%$ with acceptable computation cost.

We consider distributed optimization methods for problems where forming the Hessian is computationally challenging and communication is a significant bottleneck. We leverage randomized sketches for reducing the problem dimensions as well as preserving privacy and improving straggler resilience in asynchronous distributed systems. We derive novel approximation guarantees for classical sketching methods and establish tight concentration results that serve as both upper and lower bounds on the error. We then extend our analysis to the accuracy of parameter averaging for distributed sketches. Furthermore, we develop unbiased parameter averaging methods for randomized second order optimization for regularized problems that employ sketching of the Hessian. Existing works do not take the bias of the estimators into consideration, which limits their application to massively parallel computation. We provide closed-form formulas for regularization parameters and step sizes that provably minimize the bias for sketched Newton directions. Additionally, we demonstrate the implications of our theoretical findings via large scale experiments on a serverless cloud computing platform., arXiv admin note: text overlap with arXiv:2002.06540

In the setting of federated optimization, where a global model is aggregated periodically, step asynchronism occurs when participants conduct model training by efficiently utilizing their computational resources. It is well acknowledged that step asynchronism leads to objective inconsistency under non-i.i.d. data, which degrades the model's accuracy. To address this issue, we propose a new algorithm FedaGrac, which calibrates the local direction to a predictive global orientation. Taking advantage of the estimated orientation, we guarantee that the aggregated model does not excessively deviate from the global optimum while fully utilizing the local updates of faster nodes. We theoretically prove that FedaGrac holds an improved order of convergence rate than the state-of-the-art approaches and eliminates the negative effect of step asynchronism. Empirical results show that our algorithm accelerates the training and enhances the final accuracy., Comment: Accepted by IEEE Transactions on Parallel and Distributed Systems

Foundation models are becoming the dominant deep learning technologies. Pretraining a foundation model is always time-consumed due to the large scale of both the model parameter and training dataset. Besides being computing-intensive, the training process is extremely memory-intensive and communication-intensive. These features make it necessary to apply 3D parallelism, which integrates data parallelism, pipeline model parallelism and tensor model parallelism, to achieve high training efficiency. To achieve this goal, some custom software frameworks such as Megatron-LM and DeepSpeed are developed. However, current 3D parallelism frameworks still meet two issues: i) they are not transparent to model developers, which need to manually modify the model to parallelize training. ii) their utilization of computation, GPU memory and network bandwidth are not sufficient. We propose Merak, an automated 3D parallelism deep learning training framework with high resource utilization. Merak automatically deploys with an automatic model partitioner, which uses a graph sharding algorithm on a proxy representation of the model. Merak also presents the non-intrusive API for scaling out foundation model training with minimal code modification. In addition, we design a high-performance 3D parallel runtime engine in Merak. It uses several techniques to exploit available training resources, including shifted critical path pipeline schedule that brings a higher computation utilization, stage-aware recomputation that makes use of idle worker memory, and sub-pipelined tensor model parallelism that overlaps communication and computation. Experiments on 64 GPUs show Merak can speedup the training performance over the state-of-the-art 3D parallelism frameworks of models with 1.5, 2.5, 8.3, and 20 billion parameters by up to 1.42X, 1.39X, 1.43X, and 1.61X, respectively.

In a multiple access channel, autonomous stations are able to transmit and listen to a shared device. A fundamental problem, called \textit{contention resolution}, is to allow any station to successfully deliver its message by resolving the conflicts that arise when several stations transmit simultaneously. Despite a long history on such a problem, most of the results deal with the static setting when all stations start simultaneously, while many fundamental questions remain open in the realistic scenario when stations can join the channel at arbitrary times. In this paper, we explore the impact that three major channel features (asynchrony among stations, knowledge of the number of contenders and possibility of switching off stations after a successful transmission) can have on the time complexity of non-adaptive deterministic algorithms. We establish upper and lower bounds allowing to understand which parameters permit time-efficient contention resolution and which do not.

Many emerging cyber-physical systems, such as autonomous vehicles and robots, rely heavily on artificial intelligence and machine learning algorithms to perform important system operations. Since these highly parallel applications are computationally intensive, they need to be accelerated by graphics processing units (GPUs) to meet stringent timing constraints. However, despite the wide adoption of GPUs, efficiently scheduling multiple GPU applications while providing rigorous real-time guarantees remains a challenge. In this paper, we propose RTGPU, which can schedule the execution of multiple GPU applications in real-time to meet hard deadlines. Each GPU application can have multiple CPU execution and memory copy segments, as well as GPU kernels. We start with a model to explicitly account for the CPU and memory copy segments of these applications. We then consider the GPU architecture in the development of a precise timing model for the GPU kernels and leverage a technique known as persistent threads to implement fine-grained kernel scheduling with improved performance through interleaved execution. Next, we propose a general method for scheduling parallel GPU applications in real time. Finally, to schedule multiple parallel GPU applications, we propose a practical real-time scheduling algorithm based on federated scheduling and grid search (for GPU kernel segments) with uniprocessor fixed priority scheduling (for multiple CPU and memory copy segments). Our approach provides superior schedulability compared with previous work, and gives real-time guarantees to meet hard deadlines for multiple GPU applications according to comprehensive validation and evaluation on a real NVIDIA GTX1080Ti GPU system.

The blockchain technology has been extensively studied to enable distributed and tamper-proof data processing in federated learning (FL). Most existing blockchain assisted FL (BFL) frameworks have employed a third-party blockchain network to decentralize the model aggregation process. However, decentralized model aggregation is vulnerable to pooling and collusion attacks from the third-party blockchain network. Driven by this issue, we propose a novel BFL framework that features the integration of training and mining at the client side. To optimize the learning performance of FL, we propose to maximize the long-term time average (LTA) training data size under a constraint of LTA energy consumption. To this end, we formulate a joint optimization problem of training client selection and resource allocation (i.e., the transmit power and computation frequency at the client side), and solve the long-term mixed integer non-linear programming based on a Lyapunov technique. In particular, the proposed dynamic resource allocation and client scheduling (DRACS) algorithm can achieve a trade-off of [$\mathcal{O}(1/V)$, $\mathcal{O}(\sqrt{V})$] to balance the maximization of the LTA training data size and the minimization of the LTA energy consumption with a control parameter $V$. Our experimental results show that the proposed DRACS algorithm achieves better learning accuracy than benchmark client scheduling strategies with limited time or energy consumption., Accepted by IEEE TWC

Fog/Edge computing is a novel computing paradigm supporting resource-constrained Internet of Things (IoT) devices by the placement of their tasks on the edge and/or cloud servers. Recently, several Deep Reinforcement Learning (DRL)-based placement techniques have been proposed in fog/edge computing environments, which are only suitable for centralized setups. The training of well-performed DRL agents requires manifold training data while obtaining training data is costly. Hence, these centralized DRL-based techniques lack generalizability and quick adaptability, thus failing to efficiently tackle application placement problems. Moreover, many IoT applications are modeled as Directed Acyclic Graphs (DAGs) with diverse topologies. Satisfying dependencies of DAG-based IoT applications incur additional constraints and increase the complexity of placement problems. To overcome these challenges, we propose an actor-critic-based distributed application placement technique, working based on the IMPortance weighted Actor-Learner Architectures (IMPALA). IMPALA is known for efficient distributed experience trajectory generation that significantly reduces the exploration costs of agents. Besides, it uses an adaptive off-policy correction method for faster convergence to optimal solutions. Our technique uses recurrent layers to capture temporal behaviors of input data and a replay buffer to improve the sample efficiency. The performance results, obtained from simulation and testbed experiments, demonstrate that our technique significantly improves the execution cost of IoT applications up to 30\% compared to its counterparts., Comment: This Paper is accepted in IEEE Transactions on Mobile Computing (TMC), on 23 October 2021

There is increasing demand for specialized hardware for training deep neural networks, both in edge/IoT environments and in high-performance computing systems. The design space of such hardware is very large due to the wide range of processing architectures, deep neural network configurations, and dataflow options. This makes developing deep neural network processors quite complex, especially for training. We present TRIM, an infrastructure to help hardware architects explore the design space of deep neural network accelerators for both inference and training in the early design stages. The model evaluates at the whole network level, considering both inter-layer and intra-layer activities. Given applications, essential hardware specifications, and a design goal, TRIM can quickly explore different hardware design options, select the optimal dataflow and guide new hardware architecture design. We validated TRIM with FPGA-based implementation of deep neural network accelerators and ASIC-based architectures. We also show how to use TRIM to explore the design space through several case studies. TRIM is a powerful tool to help architects evaluate different hardware choices to develop efficient inference and training architecture design.

Federated optimization (FedOpt), which targets at collaboratively training a learning model across a large number of distributed clients, is vital for federated learning. The primary concerns in FedOpt can be attributed to the model divergence and communication efficiency, which significantly affect the performance. In this paper, we propose a new method, i.e., LoSAC, to learn from heterogeneous distributed data more efficiently. Its key algorithmic insight is to locally update the estimate for the global full gradient after {each} regular local model update. Thus, LoSAC can keep clients' information refreshed in a more compact way. In particular, we have studied the convergence result for LoSAC. Besides, the bonus of LoSAC is the ability to defend the information leakage from the recent technique Deep Leakage Gradients (DLG). Finally, experiments have verified the superiority of LoSAC comparing with state-of-the-art FedOpt algorithms. Specifically, LoSAC significantly improves communication efficiency by more than $100\%$ on average, mitigates the model divergence problem and equips with the defense ability against DLG., Comment: ACM Transactions on Knowledge Discovery from Datahttps://doi.org/10.1145/3566128

In this paper, we study a data caching problem in the cloud environment, where multiple frequently co-utilised data items could be packed as a single item being transferred to serve a sequence of data requests dynamically with reduced cost. To this end, we propose an online algorithm with respect to a homogeneous cost model, called PackCache, that can leverage the FP-Tree technique to mine those frequently co-utilised data items for packing whereby the incoming requests could be cost-effectively served online by exploiting the concept of anticipatory caching. We show the algorithm is 2\alpha competitive, reaching the lower bound of the competitive ratio for any deterministic online algorithm on the studied caching problem, and also time and space efficient to serve the requests. Finally, we evaluate the performance of the algorithm via experimental studies to show its actual cost-effectiveness and scalability., Comment: Accepted by IEEE Transactions on Computers

Unified Virtual Memory (UVM) relieves the developers from the onus of maintaining complex data structures and explicit data migration by enabling on-demand data movement between CPU memory and GPU memory. However, on-demand paging soon becomes a performance bottleneck of UVM due to the high latency caused by page table walks and data migration over interconnect. Prefetching is considered a promising solution to this problem given its ability to leverage the locality of program memory access patterns. However, existing locality-based prefetching schemes can not handle all the situations. %Data structures like arrays tend to be stored in contiguous blocks, and accessed repeatedly. An ideal prefetcher should not only look at narrow regions of the requested address space but also capture global context to deliver a good prediction of the memory access pattern. This paper proposes a novel approach for page prefetching for UVM through deep learning. We first show that a powerful Transformer learning model can provide high accuracy for UVM page prefetching. We then perform analysis to interpret this Transformer model and derive several insights that allow us to design a simpler model to match the unconstrained model's accuracy with orders of magnitude lower cost. We evaluate this simplified model on a set of 11 memory-intensive benchmarks from popular benchmark suites. Our solution outperforms the state-of-the-art UVM framework, improving the performance by 10.89%, improving the device memory page hit rate by 16.98% (89.02% vs. 76.10% for prior art), and reducing the CPU-GPU interconnect traffic by 11.05%. According to our proposed unified metric, which combines the accuracy, coverage, and page hit rate, our solution is approaching the ideal prefetching scheme more than the state-of-the-art design (0.90 vs. 0.85, with the perfect prefetcher of 1.0).

Distributed linear algebraic equation over networks, where nodes hold a part of problem data and cooperatively solve the equation via node-to-node communications, is a basic distributed computation task receiving an increasing research attention. Communications over a network have a stochastic nature, with both temporal and spatial dependence due to link failures, packet dropouts or node recreation, etc. In this paper, we study the convergence and convergence rate of distributed linear equation protocols over a $\ast$-mixing random network, where the temporal and spatial dependencies between the node-to-node communications are allowed. When the network linear equation admits exact solutions, we prove the mean-squared exponential convergence rate of the distributed projection consensus algorithm, while the lower and upper bound estimations of the convergence rate are also given for independent and identically distributed (i.i.d.) random graphs. Motivated by the randomized Kaczmarz algorithm, we also propose a distributed randomized projection consensus algorithm, where each node randomly selects one row of local linear equations for projection per iteration, and establish an exponential convergence rate. When the network linear equation admits no exact solution, we prove that a distributed gradient-descent-like algorithm with diminishing step-sizes can drive all nodes' states to a least-squares solution at a sublinear rate. These results collectively illustrate that distributed computations may overcome communication correlations if the prototype algorithms enjoy certain contractive properties or are designed with suitable parameters., 17 pages

Krylov methods are a key way of solving large sparse linear systems of equations, but suffer from poor strong scalabilty on distributed memory machines. This is due to high synchronization costs from large numbers of collective communication calls alongside a low computational workload. Enlarged Krylov methods address this issue by decreasing the total iterations to convergence, an artifact of splitting the initial residual and resulting in operations on block vectors. In this paper, we present a performance study of an Enlarged Krylov Method, Enlarged Conjugate Gradients (ECG), noting the impact of block vectors on parallel performance at scale. Most notably, we observe the increased overhead of point-to-point communication as a result of denser messages in the sparse matrix-block vector multiplication kernel. Additionally, we present models to analyze expected performance of ECG, as well as, motivate design decisions. Most importantly, we introduce a new point-to-point communication approach based on node-aware communication techniques that increases efficiency of the method at scale., 24 pages, 21 figures

Since emerging edge applications such as Internet of Things (IoT) analytics and augmented reality have tight latency constraints, hardware AI accelerators have been recently proposed to speed up deep neural network (DNN) inference run by these applications. Resource-constrained edge servers and accelerators tend to be multiplexed across multiple IoT applications, introducing the potential for performance interference between latency-sensitive workloads. In this article, we design analytic models to capture the performance of DNN inference workloads on shared edge accelerators, such as GPU and edgeTPU, under different multiplexing and concurrency behaviors. After validating our models using extensive experiments, we use them to design various cluster resource management algorithms to intelligently manage multiple applications on edge accelerators while respecting their latency constraints. We implement a prototype of our system in Kubernetes and show that our system can host 2.3× more DNN applications in heterogeneous multi-tenant edge clusters with no latency violations when compared to traditional knapsack hosting algorithms.

The Industrial Internet of Things (IIoT) is a developing research area with potential global Internet connectivity, turning everyday objects into intelligent devices with more autonomous activities. IIoT services and applications are not only being used in smart homes and smart cities, but they have also become an essential element of the Industry 4.0 concept. The emergence of the IIoT helps traditional industries simplify production processes, reduce production costs, and improve industrial efficiency. However, the involvement of many heterogeneous devices, the use of third-party software, and the resource-constrained nature of the IoT devices bring new security risks to the production chain and expose vulnerabilities to the systems. The Distributed Denial of Service (DDoS) attacks are significant, among others. This article analyzes the threats and attacks in the IIoT and discusses how DDoS attacks impact the production process and communication dysfunctions with IIoT services and applications. This article also proposes a reference security framework that enhances the advantages of fog computing to demonstrate countermeasures against DDoS attacks and possible strategies to mitigate such attacks at scale.

Edge intelligence has arisen as a promising computing paradigm for supporting miscellaneous smart applications that rely on machine learning techniques. While the community has extensively investigated multi-tier edge deployment for traditional deep learning models (e.g. CNNs, RNNs), the emerging Graph Neural Networks (GNNs) are still under exploration, presenting a stark disparity to its broad edge adoptions such as traffic flow forecasting and location-based social recommendation. To bridge this gap, this paper formally studies the cost optimization for distributed GNN processing over a multi-tier heterogeneous edge network. We build a comprehensive modeling framework that can capture a variety of different cost factors, based on which we formulate a cost-efficient graph layout optimization problem that is proved to be NP-hard. Instead of trivially applying traditional data placement wisdom, we theoretically reveal the structural property of quadratic submodularity implicated in GNN's unique computing pattern, which motivates our design of an efficient iterative solution exploiting graph cuts. Rigorous analysis shows that it provides parameterized constant approximation ratio, guaranteed convergence, and exact feasibility. To tackle potential graph topological evolution in GNN processing, we further devise an incremental update strategy and an adaptive scheduling algorithm for lightweight dynamic layout optimization. Evaluations with real-world datasets and various GNN benchmarks demonstrate that our approach achieves superior performance over de facto baselines with more than 95.8% cost eduction in a fast convergence speed., Comment: 19 pages, 20 figures, accepted by IEEE Journal of Selected Areas in Communications

The amount of CO$_2$ emitted per kilowatt-hour on an electricity grid varies by time of day and substantially varies by location due to the types of generation. Networked collections of warehouse scale computers, sometimes called Hyperscale Computing, emit more carbon than needed if operated without regard to these variations in carbon intensity. This paper introduces Google's system for Carbon-Intelligent Compute Management, which actively minimizes electricity-based carbon footprint and power infrastructure costs by delaying temporally flexible workloads. The core component of the system is a suite of analytical pipelines used to gather the next day's carbon intensity forecasts, train day-ahead demand prediction models, and use risk-aware optimization to generate the next day's carbon-aware Virtual Capacity Curves (VCCs) for all datacenter clusters across Google's fleet. VCCs impose hourly limits on resources available to temporally flexible workloads while preserving overall daily capacity, enabling all such workloads to complete within a day. Data from operation shows that VCCs effectively limit hourly capacity when the grid's energy supply mix is carbon intensive and delay the execution of temporally flexible workloads to "greener" times.

Moving beyond just interconnected devices, the increasing interplay between communication and computation has fed the vision of real-time networked control systems. To obtain timely situational awareness, IoT devices continuously sample computation-intensive status updates, generate perception tasks and offload them to edge servers for processing. In this sense, the timeliness of information is considered as one major contextual attribute of status updates. In this paper, we derive the closed-form expressions of timeliness of information for computation offloading at both edge tier and fog tier, where two stage tandem queues are exploited to abstract the transmission and computation process. Moreover, we exploit the statistical structure of Gauss-Markov process, which is widely adopted to model temporal dynamics of system states, and derive the closed-form expression for process-related timeliness of information. The obtained analytical formulas explicitly characterize the dependency among task generation, transmission and execution, which can serve as objective functions for system optimization. Based on the theoretical results, we formulate a computation offloading optimization problem at edge tier, where the timeliness of status updates is minimized among multiple devices by joint optimization of task generation, bandwidth allocation, and computation resource allocation. An iterative solution procedure is proposed to solve the formulated problem. Numerical results reveal the intertwined relationship among transmission and computation stages, and verify the necessity of factoring in the task generation process for computation offloading strategy design.

Massive Internet of Things (IoT) networks have a wide range of applications, including but not limited to the rapid delivery of emergency and disaster messages. Although various benchmark algorithms have been developed to date for message delivery in such applications, they pose several practical challenges such as insufficient network coverage and/or highly redundant transmissions to expand the coverage area, resulting in considerable energy consumption for each IoT device. To overcome this problem, we first characterize a new performance metric, forwarding efficiency, which is defined as the ratio of the coverage probability to the average number of transmissions per device, to evaluate the data dissemination performance more appropriately. Then, we propose a novel and effective forwarding method, fishbone forwarding (FiFo), which aims to improve the forwarding efficiency with acceptable computational complexity. Our FiFo method completes two tasks: 1) it clusters devices based on the unweighted pair group method with the arithmetic average; and 2) it creates the main axis and sub axes of each cluster using both the expectation-maximization algorithm for the Gaussian mixture model and principal component analysis. We demonstrate the superiority of FiFo by using a real-world dataset. Through intensive and comprehensive simulations, we show that the proposed FiFo method outperforms benchmark algorithms in terms of the forwarding efficiency., Comment: 13 pages, 16 figures, 5 tables; to appear in the IEEE Internet of Things Journal (Please cite our journal version that will appear in an upcoming issue.)

With the proliferation of smart devices and revolutions in communications, electrical distribution systems are gradually shifting from passive, manually-operated and inflexible ones, to a massively interconnected cyber-physical smart grid to address the energy challenges of the future. However, the integration of several cutting-edge technologies has introduced several security and privacy vulnerabilities due to the large-scale complexity and resource limitations of deployments. Recent research trends have shown that False Data Injection (FDI) attacks are becoming one of the most malicious cyber threats within the entire smart grid paradigm. Therefore, this paper presents a comprehensive survey of the recent advances in FDI attacks within active distribution systems and proposes a taxonomy to classify the FDI threats with respect to smart grid targets. The related studies are contrasted and summarized in terms of the attack methodologies and implications on the electrical power distribution networks. Finally, we identify some research gaps and recommend a number of future research directions to guide and motivate prospective researchers.

The orbital propagation of large sets of initial conditions under high accuracy requirements is currently a bottleneck in the development of space missions, e.g. for planetary protection compliance analyses. The proposed approach can include any force source in the dynamical model through efficient Picard-Chebyshev (PC) numerical simulations. A two-level augmentation of the integration scheme is proposed, to run an arbitrary number of simulations within the same algorithm call, fully exploiting high performance and GPU (Graphics Processing Units) computing facilities. The performances obtained with implementation in C and NVIDIA CUDA programming languages are shown, on a test case taken from the optimization of a Solar Orbiter-like first resonant phase with Venus.

Vehicle tracking, a core application to smart city video analytics, is becoming more widely deployed than ever before thanks to the increasing number of traffic cameras and recent advances in computer vision and machine-learning. Due to the constraints of bandwidth, latency, and privacy concerns, tracking tasks are more preferable to run on edge devices sitting close to the cameras. However, edge devices are provisioned with a fixed amount of computing budget, making them incompetent to adapt to time-varying and imbalanced tracking workloads caused by traffic dynamics. In coping with this challenge, we propose WatchDog, a real-time vehicle tracking system that fully utilizes edge nodes across the road network. WatchDog leverages computer vision tasks with different resource-accuracy tradeoffs, and decomposes and schedules tracking tasks judiciously across edge devices based on the current workload to maximize the number of tasks while ensuring a provable response time-bound at each edge device. Extensive evaluations have been conducted using real-world city-wide vehicle trajectory datasets, achieving exceptional tracking performance with a real-time guarantee.

The advent of Federated Learning (FL) has ignited a new paradigm for parallel and confidential decentralized Machine Learning (ML) with the potential of utilizing the computational power of a vast number of IoT, mobile and edge devices without data leaving the respective device, ensuring privacy by design. Yet, in order to scale this new paradigm beyond small groups of already entrusted entities towards mass adoption, the Federated Learning Framework (FLF) has to become (i) truly decentralized and (ii) participants have to be incentivized. This is the first systematic literature review analyzing holistic FLFs in the domain of both, decentralized and incentivized federated learning. 422 publications were retrieved, by querying 12 major scientific databases. Finally, 40 articles remained after a systematic review and filtering process for in-depth examination. Although having massive potential to direct the future of a more distributed and secure AI, none of the analyzed FLF is production-ready. The approaches vary heavily in terms of use-cases, system design, solved issues and thoroughness. We are the first to provide a systematic approach to classify and quantify differences between FLF, exposing limitations of current works and derive future directions for research in this novel domain., submitted to IEEE IOTJ

We consider a generalization of the recently proposed gradient coding framework where a large dataset is divided across $n$ workers and each worker transmits to a master node one or more linear combinations of the gradients over the data subsets assigned to it. Unlike the conventional framework which requires the master node to recover the sum of the gradients over all the data subsets in the presence of $s$ straggler workers, we relax the goal of the master node to computing the sum of at least some α fraction of the gradients. The broad goal of our work is to study the optimal computation and communication load per worker for this approximate gradient coding framework. We begin by deriving a lower bound on the computation load of any feasible scheme and also propose a strategy which achieves this lower bound, albeit at the cost of high communication load and a number of data partitions which can be polynomial in the number of workers n. We then restrict attention to schemes which utilize a number of data partitions equal to $n$ and propose schemes based on cyclic assignment which have a lower communication load. When each worker transmits a single linear combination, we also prove lower bounds on the computation load of any scheme using $n$ data partitions. A full version of this paper is accessible at: https://arxiv.org/abs/2102.10163

In this paper, we investigate the architecture of an optimal controller that maximizes the convergence speed of a consensus protocol with single-integrator dynamics. Under the assumption that communication delays increase with the number of hops from which information is allowed to reach each agent, we address the optimal control design under delayed feedback and show that the optimal controller features, in general, a sparsely connected architecture., 6 pages, 7 figures; published on IEEE Control Systems Letters

K-nearest neighbor search is one of the fundamental tasks in various applications and the hierarchical navigable small world (HNSW) has recently drawn attention in large-scale cloud services, as it easily scales up the database while offering fast search. On the other hand, a computational storage device (CSD) that combines programmable logic and storage modules on a single board becomes popular to address the data bandwidth bottleneck of modern computing systems. In this paper, we propose a computational storage platform that can accelerate a large-scale graph-based nearest neighbor search algorithm based on SmartSSD CSD. To this end, we modify the algorithm more amenable on the hardware and implement two types of accelerators using HLS- and RTL-based methodology with various optimization methods. In addition, we scale up the proposed platform to have 4 SmartSSDs and apply graph parallelism to boost the system performance further. As a result, the proposed computational storage platform achieves 75.59 query per second throughput for the SIFT1B dataset at 258.66W power dissipation, which is 12.83x and 17.91x faster and 10.43x and 24.33x more energy efficient than the conventional CPU-based and GPU-based server platform, respectively. With multi-terabyte storage and custom acceleration capability, we believe that the proposed computational storage platform is a promising solution for cost-sensitive cloud datacenters., Extension of FCCM 20201 and Accepted in Transaction on Computers

In recent years, deep learning (DL) models have demonstrated remarkable achievements on non-trivial tasks such as speech recognition and natural language understanding. One of the significant contributors to its success is the proliferation of end devices that acted as a catalyst to provide data for data-hungry DL models. However, computing DL training and inference is the main challenge. Usually, central cloud servers are used for the computation, but it opens up other significant challenges, such as high latency, increased communication costs, and privacy concerns. To mitigate these drawbacks, considerable efforts have been made to push the processing of DL models to edge servers. Moreover, the confluence point of DL and edge has given rise to edge intelligence (EI). This survey paper focuses primarily on the fifth level of EI, called all in-edge level, where DL training and inference (deployment) are performed solely by edge servers. All in-edge is suitable when the end devices have low computing resources, e.g., Internet-of-Things, and other requirements such as latency and communication cost are important in mission-critical applications, e.g., health care. Firstly, this paper presents all in-edge computing architectures, including centralized, decentralized, and distributed. Secondly, this paper presents enabling technologies, such as model parallelism and split learning, which facilitate DL training and deployment at edge servers. Thirdly, model adaptation techniques based on model compression and conditional computation are described because the standard cloud-based DL deployment cannot be directly applied to all in-edge due to its limited computational resources. Fourthly, this paper discusses eleven key performance metrics to evaluate the performance of DL at all in-edge efficiently. Finally, several open research challenges in the area of all in-edge are presented., 21 pages

Federated learning (FL) has been promoted as a popular technique for training machine learning (ML) models over edge/fog networks. Traditional implementations of FL have largely neglected the potential for inter-network cooperation, treating edge/fog devices and other infrastructure participating in ML as separate processing elements. Consequently, FL has been vulnerable to several dimensions of network heterogeneity, such as varying computation capabilities, communication resources, data qualities, and privacy demands. We advocate for cooperative federated learning (CFL), a cooperative edge/fog ML paradigm built on device-to-device (D2D) and device-to-server (D2S) interactions. Through D2D and D2S cooperation, CFL counteracts network heterogeneity in edge/fog networks through enabling a model/data/resource pooling mechanism, which will yield substantial improvements in ML model training quality and network resource consumption. We propose a set of core methodologies that form the foundation of D2D and D2S cooperation and present preliminary experiments that demonstrate their benefits. We also discuss new FL functionalities enabled by this cooperative framework such as the integration of unlabeled data and heterogeneous device privacy into ML model training. Finally, we describe some open research directions at the intersection of cooperative edge/fog and FL., This paper has been accepted for publication in IEEE Communications Magazine

We propose a flexible low complexity design (FLCD) of coded distributed computing (CDC) with empirical evaluation on Amazon Elastic Compute Cloud (Amazon EC2). CDC can expedite MapReduce like computation by trading increased map computations to reduce communication load and shuffle time. A main novelty of FLCD is to utilize the design freedom in defining map and reduce functions to develop asymptotic homogeneous systems to support varying intermediate values (IV) sizes under a general MapReduce framework. Compared to existing designs with constant IV sizes, FLCD offers greater flexibility in adapting to network parameters and significantly reduces the implementation complexity by requiring fewer input files and shuffle groups. The FLCD scheme is the first proposed low-complexity CDC design that can operate on a network with an arbitrary number of nodes and computation load. We perform empirical evaluations of the FLCD by executing the TeraSort algorithm on an Amazon EC2 cluster. This is the first time that theoretical predictions of the CDC shuffle time are validated by empirical evaluations. The evaluations demonstrate a 2.0 to 4.24x speedup compared to conventional uncoded MapReduce, a 12% to 52% reduction in total time, and a wider range of operating network parameters compared to existing CDC schemes., 13 pages, 4 figures

Both Byzantine resilience and communication efficiency have attracted tremendous attention recently for their significance in edge federated learning. However, most existing algorithms may fail when dealing with real-world irregular data that behaves in a heavy-tailed manner. To address this issue, we study the stochastic convex and non-convex optimization problem for federated learning at edge and show how to handle heavy-tailed data while retaining the Byzantine resilience, communication efficiency and the optimal statistical error rates simultaneously. Specifically, we first present a Byzantine-resilient distributed gradient descent algorithm that can handle the heavy-tailed data and meanwhile converge under the standard assumptions. To reduce the communication overhead, we further propose another algorithm that incorporates gradient compression techniques to save communication costs during the learning process. Theoretical analysis shows that our algorithms achieve order-optimal statistical error rate in presence of Byzantine devices. Finally, we conduct extensive experiments on both synthetic and real-world datasets to verify the efficacy of our algorithms.

Given the proximity of many wireless users and their diversity in consuming local resources (e.g., data-plans, computation and energy resources), device-to-device (D2D) resource sharing is a promising approach towards realizing a sharing economy. This paper adopts an easy-to-implement greedy matching algorithm with distributed fashion and only sub-linear O(log n) parallel complexity (in user number n) for large-scale D2D sharing. Practical cases indicate that the greedy matching's average performance is far better than the worst-case approximation ratio 50% as compared to the optimum. However, there is no rigorous average-case analysis in the literature to back up such encouraging findings and this paper is the first to present such analysis for multiple representative classes of graphs. For 1D linear networks, we prove that our greedy algorithm performs better than 86.5% of the optimum. For 2D grids, though dynamic programming cannot be directly applied, we still prove this average performance ratio to be above 76%. For the more challenging Erdos-Renyi random graphs, we equivalently reduce to the asymptotic analysis of random trees and successfully prove a ratio up to 79%. Finally, we conduct experiments using real data to simulate realistic D2D networks, and show that our analytical performance measure approximates well practical cases., Comment: Accepted by IEEE Transactions on Mobile Computing. arXiv admin note: substantial text overlap with arXiv:2107.12581

Machine learning (ML) is a widely accepted means for supporting customized services for mobile devices and applications. Federated Learning (FL), which is a promising approach to implement machine learning while addressing data privacy concerns, typically involves a large number of wireless mobile devices to collect model training data. Under such circumstances, FL is expected to meet stringent training latency requirements in the face of limited resources such as demand for wireless bandwidth, power consumption, and computation constraints of participating devices. Due to practical considerations, FL selects a portion of devices to participate in the model training process at each iteration. Therefore, the tasks of efficient resource management and device selection will have a significant impact on the practical uses of FL. In this paper, we propose a spectrum allocation optimization mechanism for enhancing FL over a wireless mobile network. Specifically, the proposed spectrum allocation optimization mechanism minimizes the time delay of FL while considering the energy consumption of individual participating devices; thus ensuring that all the participating devices have sufficient resources to train their local models. In this connection, to ensure fast convergence of FL, a robust device selection is also proposed to help FL reach convergence swiftly, especially when the local datasets of the devices are not independent and identically distributed (non-iid). Experimental results show that (1) the proposed spectrum allocation optimization method optimizes time delay while satisfying the individual energy constraints; (2) the proposed device selection method enables FL to achieve the fastest convergence on non-iid datasets., Comment: This paper is accepted by IEEE/ACM Transactions on Networking

In this paper, we consider distributed optimization problems where $n$ agents, each possessing a local cost function, collaboratively minimize the average of the local cost functions over a connected network. To solve the problem, we propose a distributed random reshuffling (D-RR) algorithm that invokes the random reshuffling (RR) update in each agent. We show that D-RR inherits favorable characteristics of RR for both smooth strongly convex and smooth nonconvex objective functions. In particular, for smooth strongly convex objective functions, D-RR achieves $\mathcal{O}(1/T^2)$ rate of convergence (where $T$ counts epoch number) in terms of the squared distance between the iterate and the global minimizer. When the objective function is assumed to be smooth nonconvex, we show that D-RR drives the squared norm of gradient to $0$ at a rate of $\mathcal{O}(1/T^{2/3})$. These convergence results match those of centralized RR (up to constant factors) and outperform the distributed stochastic gradient descent (DSGD) algorithm if we run a relatively large number of epochs. Finally, we conduct a set of numerical experiments to illustrate the efficiency of the proposed D-RR method on both strongly convex and nonconvex distributed optimization problems., 20 pages, 13 figures

Hyperledger Fabric (HLF), one of the most popular private blockchains, has recently received attention for blockchain-enabled Internet of Things (IoT). However, for IoT applications to handle time-sensitive data, the processing latency in HLF has emerged as a new challenge. In this article, therefore, we establish a practical HLF latency model for HLF-enabled IoT. We first discuss the structure and transaction flow of HLF-enabled IoT. After implementing real HLF, we capture the latencies that each transaction experiences and show that the total latency of HLF can be modeled as a Gamma distribution, which is validated by conducting a goodness-of-fit test (i.e., the Kolmogorov-Smirnov (KS) test). We also provide the parameter values of the modeled latency distribution for various HLF environments. Furthermore, we explore the impacts of three important HLF parameters including transaction generation rate, block size, and block-generation timeout on the HLF latency. As a result, this article provides design insights on minimizing the latency for HLF-enabled IoT., Comment: 7 pages, 5 figures, 1 table

In this paper, we propose a Similarity-Quality-based Messenger Distillation (SQMD) framework for heterogeneous asynchronous on-device healthcare analytics. By introducing a preloaded reference dataset, SQMD enables all participant devices to distill knowledge from peers via messengers (i.e., the soft labels of the reference dataset generated by clients) without assuming the same model architecture. Furthermore, the messengers also carry important auxiliary information to calculate the similarity between clients and evaluate the quality of each client model, based on which the central server creates and maintains a dynamic collaboration graph (communication graph) to improve the personalization and reliability of SQMD under asynchronous conditions. Extensive experiments on three real-life datasets show that SQMD achieves superior performance.

Spreading information through a network of devices is a core activity for most distributed systems. As such, self-stabilizing algorithms implementing information spreading are one of the key building blocks enabling aggregate computing to provide resilient coordination in open complex distributed systems. This paper improves a general spreading block in the aggregate computing literature by making it resilient to network perturbations, establishes its global uniform asymptotic stability and proves that it is ultimately bounded under persistent disturbances. The ultimate bounds depend only on the magnitude of the largest perturbation and the network diameter, and three design parameters trade off competing aspects of performance. For example, as in many dynamical systems, values leading to greater resilience to network perturbations slow convergence and vice versa.

Federated learning (FL) is an emerging machine learning paradigm that allows to accomplish model training without aggregating data at a central server. Most studies on FL consider a centralized framework, in which a single server is endowed with a central authority to coordinate a number of devices to perform model training in an iterative manner. Due to stringent communication and bandwidth constraints, such a centralized framework has limited scalability as the number of devices grows. To address this issue, in this paper, we propose a ConFederated Learning (CFL) framework. The proposed CFL consists of multiple servers, in which each server is connected with an individual set of devices as in the conventional FL framework, and decentralized collaboration is leveraged among servers to make full use of the data dispersed throughout the network. We develop an alternating direction method of multipliers (ADMM) algorithm for CFL. The proposed algorithm employs a random scheduling policy which randomly selects a subset of devices to access their respective servers at each iteration, thus alleviating the need of uploading a huge amount of information from devices to servers. Theoretical analysis is presented to justify the proposed method. Numerical results show that the proposed method can converge to a decent solution significantly faster than gradient-based FL algorithms, thus boasting a substantial advantage in terms of communication efficiency., 13 pages, 5 figures

Everyday, large amounts of sensitive data is distributed across mobile phones, wearable devices, and other sensors. Traditionally, these enormous datasets have been processed on a single system, with complex models being trained to make valuable predictions. Distributed machine learning techniques such as Federated and Split Learning have recently been developed to protect user data and privacy better while ensuring high performance. Both of these distributed learning architectures have advantages and disadvantages. In this paper, we examine these tradeoffs and suggest a new hybrid Federated Split Learning architecture that combines the efficiency and privacy benefits of both. Our evaluation demonstrates how our hybrid Federated Split Learning approach can lower the amount of processing power required by each client running a distributed learning system, reduce training and inference time while keeping a similar accuracy. We also discuss the resiliency of our approach to deep learning privacy inference attacks and compare our solution to other recently proposed benchmarks.

The vehicular edge computing (VEC) can cache contents in different RSUs at the network edge to support the real-time vehicular applications. In VEC, owing to the high-mobility characteristics of vehicles, it is necessary to cache the user data in advance and learn the most popular and interesting contents for vehicular users. Since user data usually contains privacy information, users are reluctant to share their data with others. To solve this problem, traditional federated learning (FL) needs to update the global model synchronously through aggregating all users' local models to protect users' privacy. However, vehicles may frequently drive out of the coverage area of the VEC before they achieve their local model trainings and thus the local models cannot be uploaded as expected, which would reduce the accuracy of the global model. In addition, the caching capacity of the local RSU is limited and the popular contents are diverse, thus the size of the predicted popular contents usually exceeds the cache capacity of the local RSU. Hence, the VEC should cache the predicted popular contents in different RSUs while considering the content transmission delay. In this paper, we consider the mobility of vehicles and propose a cooperative Caching scheme in the VEC based on Asynchronous Federated and deep Reinforcement learning (CAFR). We first consider the mobility of vehicles and propose an asynchronous FL algorithm to obtain an accurate global model, and then propose an algorithm to predict the popular contents based on the global model. In addition, we consider the mobility of vehicles and propose a deep reinforcement learning algorithm to obtain the optimal cooperative caching location for the predicted popular contents in order to optimize the content transmission delay. Extensive experimental results have demonstrated that the CAFR scheme outperforms other baseline caching schemes., Comment: This paper has been submitted to IEEE Journal of Selected Topics in Signal Processing

We tackle the problem of a set of agents achieving resilient consensus in the presence of attacked agents. We present a discrete-time reputation-based consensus algorithm for synchronous and asynchronous networks by developing a local strategy where, at each time, each agent assigns a reputation (between zero and one) to each neighbor. The reputation is then used to weigh the neighbors' values in the update of its state. Under mild assumptions, we show that: (i) the proposed method converges exponentially to the consensus of the regular agents; (ii) if a regular agent identifies a neighbor as an attacked node, then it is indeed an attacked node; (iii) if the consensus value of the normal nodes differs from that of any of the attacked nodes' values, then the reputation that a regular agent assigns to the attacked neighbors goes to zero. Further, we extend our method to achieve resilience in the scenarios where there are noisy nodes, dynamic networks and stochastic node selection. Finally, we illustrate our algorithm with several examples, and we delineate some attacking scenarios that can be dealt by the current proposal but not by the state-of-the-art approaches.

Information compression is essential to reduce communication cost in distributed optimization over peer-to-peer networks. This paper proposes a communication-efficient linearly convergent distributed (COLD) algorithm to solve strongly convex optimization problems. By compressing innovation vectors, which are the differences between decision vectors and their estimates, COLD is able to achieve linear convergence for a class of $\delta$-contracted compressors. We explicitly quantify how the compression affects the convergence rate and show that COLD matches the same rate of its uncompressed version. To accommodate a wider class of compressors that includes the binary quantizer, we further design a novel dynamical scaling mechanism and obtain the linearly convergent Dyna-COLD. Importantly, our results strictly improve existing results for the quantized consensus problem. Numerical experiments demonstrate the advantages of both algorithms under different compressors., Comment: 14 pages

Data-driven predictive control (DPC) has been studied and used in various scenarios, since it could generate the predicted control sequence only relying on the historical input and output data. Recently, based on cloud computing, data-driven predictive cloud control system (DPCCS) has been proposed with the advantage of sufficient computational resources. However, the existing computation mode of DPCCS is centralized. This computation mode could not utilize fully the computing power of cloud computing, of which the structure is distributed. Thus, the computation delay could not been reduced and still affects the control quality. In this paper, a novel cloud-edge collaborative containerised workflow-based DPC system with disturbance observer (DOB) is proposed, to improve the computation efficiency and guarantee the control accuracy. First, a construction method for the DPC workflow is designed, to match the distributed processing environment of cloud computing. But the non-computation overheads of the workflow tasks are relatively high. Therefore, a cloud-edge collaborative control scheme with DOB is designed. The low-weight data could be truncated to reduce the non-computation overheads. Meanwhile, we design an edge DOB to estimate and compensate the uncertainty in cloud workflow processing, and obtain the composite control variable. The UUB stability of the DOB is also proved. Third, to execute the workflow-based DPC controller and evaluate the proposed cloud-edge collaborative control scheme with DOB in the real cloud environment, we design and implement a practical workflow-based cloud control experimental system based on container technology. Finally, a series of evaluations show that, the computation times are decreased by 45.19% and 74.35% for two real-time control examples, respectively, and by at most 85.10% for a high-dimension control example., Comment: 58 pages and 23 figures

In business process landscapes, a common challenge is to provide the necessary computational resources to enact the single process steps. One well-known approach to solve this issue in a cost-efficient way is to use the notion of elasticity, i.e., to provide cloud-based computational resources in a rapid fashion and to enact the single process steps on these resources. Existing approaches to provide elastic processes are mostly based on Virtual Machines (VMs). Utilizing container technologies could enable a more fine-grained allocation of process steps to computational resources, leading to a better resource utilization and improved cost efficiency. In this paper, we propose an approach to optimize resource allocation for elastic processes by applying a four-fold auto-scaling approach. The main goal is to minimize the cost of process enactments by using containers. To this end, we formulate and implement a multi-objective optimization problem applying Mixed-Integer Linear Programming and use a transformation step to allocate software services to containers. We thoroughly evaluate the optimization problem and show that it can lead to significant cost savings while maintaining Service Lev

We propose a novel context-aware drone delivery framework for optimizing package delivery through skyway networks in smart cities. We reformulate the problem of finding an optimal drone service delivery pathway as a more congruent and elegant drone delivery service composition problem. In this respect, we propose a novel line-of-sight heuristic-based context-aware composition algorithm that selects and composes near-optimal drone delivery services. We conducted an extensive experiment using a real dataset to show the robustness of our proposed approach., Comment: 8 pages, 3 figures. This is an accepted paper and it is going to appear in IEEE Internet Computing magazine

We consider a discrete system of $n$ simple indistinguishable devices, called \emph{agents}, forming a \emph{connected} shape $S_I$ on a two-dimensional square grid. Agents are equipped with a linear-strength mechanism, called a \emph{line move}, by which an agent can push a whole line of consecutive agents in one of the four directions in a single time-step. We study the problem of transforming an initial shape $S_I$ into a given target shape $S_F$ via a finite sequence of line moves in a distributed model, where each agent can observe the states of nearby agents in a Moore neighbourhood. Our main contribution is the first distributed connectivity-preserving transformation that exploits line moves within a total of $O(n \log_2 n)$ moves, which is asymptotically equivalent to that of the best-known centralised transformations. The algorithm solves the \emph{line formation problem} that allows agents to form a final straight line $S_L$, starting from any shape $ S_I $, whose \emph{associated graph} contains a Hamiltonian path., Comment: 31 pages, 18 figures

The operational cost of a cloud computing platform is one of the most significant Quality of Service (QoS) criteria for schedulers, crucial to keep up with the growing computational demands. Several data-driven deep neural network (DNN)-based schedulers have been proposed in recent years that outperform alternative approaches by providing scalable and effective resource management for dynamic workloads. However, state-of-the-art schedulers rely on advanced DNNs with high computational requirements, implying high scheduling costs. In non-stationary contexts, the most sophisticated schedulers may not always be required, and it may be sufficient to rely on low-cost schedulers to temporarily save operational costs. In this work, we propose MetaNet, a surrogate model that predicts the operational costs and scheduling overheads of a large number of DNN-based schedulers and chooses one on-the-fly to jointly optimize job scheduling and execution costs. This facilitates improvements in execution costs, energy usage and service level agreement violations of up to 11%, 43% and 13% compared to the state-of-the-art methods., Accepted as a poster in SIGMETRICS 2022

As deep neural networks continue to expand and become more complex, most edge devices are unable to handle their extensive processing requirements. Therefore, the concept of distributed inference is essential to distribute the neural network among a cluster of nodes. However, distribution may lead to additional energy consumption and dependency among devices that suffer from unstable transmission rates. Unstable transmission rates harm real-time performance of IoT devices causing low latency, high energy usage, and potential failures. Hence, for dynamic systems, it is necessary to have a resilient DNN with an adaptive architecture that can downsize as per the available resources. This paper presents an empirical study that identifies the connections in ResNet that can be dropped without significantly impacting the model's performance to enable distribution in case of resource shortage. Based on the results, a multi-objective optimization problem is formulated to minimize latency and maximize accuracy as per available resources. Our experiments demonstrate that an adaptive ResNet architecture can reduce shared data, energy consumption, and latency throughout the distribution while maintaining high accuracy., Accepted in the International Wireless Communications & Mobile Computing Conference (IWCMC 2023)

We provide an overview of the software engineering efforts and their impact in QMCPACK, a production-level ab-initio Quantum Monte Carlo open-source code targeting high-performance computing (HPC) systems. Aspects included are: (i) strategic expansion of continuous integration (CI) targeting CPUs, using GitHub Actions runners, and NVIDIA and AMD GPUs in pre-exascale systems, using self-hosted hardware; (ii) incremental reduction of memory leaks using sanitizers, (iii) incorporation of Docker containers for CI and reproducibility, and (iv) refactoring efforts to improve maintainability, testing coverage, and memory lifetime management. We quantify the value of these improvements by providing metrics to illustrate the shift towards a predictive, rather than reactive, sustainable maintenance approach. Our goal, in documenting the impact of these efforts on QMCPACK, is to contribute to the body of knowledge on the importance of research software engineering (RSE) for the sustainability of community HPC codes and scientific discovery at scale., Accepted at the first US-RSE Conference, USRSE2023, https://us-rse.org/usrse23/, 8 pages, 3 figures, 4 tables