Search

BCube is a well-known recursively defined network structure. It provides multiple low-diameter paths and good fault-tolerance for data center networks (DCNs). Its distributed routing algorithm, BCube Source Routing (BSR), can be deployed rapidly and conveniently to build multiple parallel path sets. But in the worst case, BSR may suffer from flow collisions and waste $50\%$ of the capacity of each BCube link. In this paper, to fully use those redundant links and to improve BCube load balancing, we supplement the BCube topology with a central master computer and we design two centralized dynamic parallel flow scheduling algorithms: CDPFS and CDPFSMP, for single-path and multi-path respectively. We focus on finding the least congested path for each flow by analyzing the information about the state of the global network. Furthermore, we allocate those paths to each flow in parallel. The simulation result shows that our proposed algorithms take advantage of BCube structure and deliver high-performance solutions for load balancing problems.

In recent years, edge computing has become an increasingly popular computing paradigm to enable real-time data processing and mobile intelligence. Edge computing allows computing at the edge of the network, where data is generated and distributed at the nearby edge servers to reduce the data access latency and improve data processing efficiency. One of the key challenges in data-intensive edge computing is how to place the data at the edge clouds effectively such that the access latency to the data is minimized. In this paper, we study such a data placement problem in edge computing while different data items have diverse popularity. We propose a popularity based placement method which maps both data items and edge servers to a virtual plane and places or retrieves data based on its virtual coordinate in the plane. We then further propose additional placement strategies to handle load balancing among edge servers via either offloading or data duplication. Simulation results show that our proposed strategies efficiently reduce the average path length of data access and the load-balancing strategies indeed provide an effective relief of storage pressures at certain overloaded servers.

Machine learning (ML) models are increasingly trained with distributed workers possessing heterogeneous resources. In such scenarios, model training efficiency may be negatively affected by \emph{stragglers}---workers that run much slower than others. Efficient model training requires eliminating such stragglers, yet for modern ML workloads, existing load balancing strategies are inefficient and even infeasible. In this paper, we propose a novel strategy, called \emph{semi-dynamic load balancing}, to eliminate stragglers of distributed ML workloads. The key insight is that ML workers shall be load-balanced at \emph{iteration boundaries}, being non-intrusive to intra-iteration execution. Based on it we further develop LB-BSP, an integrated worker coordination mechanism that adapts workers' load to their instantaneous processing capabilities---by right-sizing the sample batches at the synchronization barriers. We have designed distinct load tuning algorithms for ML in CPU clusters, in GPU clusters as well as in federated learning setups, based on their respective characteristics. LB-BSP has been implemented as a Python module for ML frameworks like TensorFlow and PyTorch. Our EC2 deployment confirms that LB-BSP is practical, effective and light-weight, and is able to accelerating distributed training by up to $54\%$ .

In this study, a self-learning discrete Jaya algorithm (SD-Jaya) is proposed to address the energy-efficient distributed no-idle flow-shop scheduling problem (FSP) in a heterogeneous factory system (HFS-EEDNIFSP) with the criteria of minimizing the total tardiness (TTD), total energy consumption (TEC), and factory load balancing (FLB). First, the mixed-integer programming model of HFS-EEDNIFSP is presented. An evaluation criterion of FLB combining the energy consumption and the completion time is introduced. Second, a self-learning operators selection strategy, in which the success rate of each operator is summarized as knowledge, is designed for guiding the selection of operators. Third, the energy-saving strategy is proposed for reducing the TEC. The energy-efficient no-idle FSP is transformed to be an energy-efficient permutation FSP to search the idle times. The speed of operations which adjacent are idle times is reduced. The effectiveness of SD-Jaya is tested on 60 benchmark instances. On the quality of the solution, the experimental results reveal that the efficacy of the SD-Jaya algorithm outperforms the other algorithms for addressing HFS-EEDNIFSP.

Given a dataset $\cal S$ composed of g subsets with each data items belonging to one of them, $\textit{multiset membership lookup}$ takes an item e as input and outputs a binary answer whether $e\in{\cal S}$ and, in case of yes, the ID of the subset to which e belongs. Overlaid upon while more sophisticated than the canonical membership lookup, multiset membership lookup emerges as a pivotal functionality in many computing and networking paradigms. The quest to achieve high-speed, high-accuracy lookup with limited memory cost makes lookup algorithm design a challenging task, particularly when the data items arrive as a stream. In this paper, we devise compact data structures and lookup algorithms that are amendable for hardware implementation, while guaranteeing high lookup accuracy and supporting interactive query processing. We first propose $\textit{multi-hash color table}$ , a variant of Bloom filter, to encode subset IDs compactly and map the ID of an item to its subset ID. We further construct a more balanced data structure called $\textit{balanced multi-hash color table}$ to improve the compactness by integrating the state-of-the-art load balancing technique. We complete our work by addressing the case of $\textit{batch arrivals}$ and design a batched recording algorithm optimizing the memory efficiency.

The elasticity of cloud resources allow cloud clients to expand and shrink their demand of resources dynamically over time. However, fluctuations in the resource demands and pre-defined size of virtual machines (VMs) lead to lack of resource utilization, load imbalance and excessive power consumption. To address these issues and to improve the performance of datacenter, an efficient resource management framework is proposed, which anticipates resource utilization of the servers and balances the load accordingly. It facilitates power saving, by minimizing the number of active servers, VM migrations and maximizing the resource utilization. An online resource prediction system, is developed and installed at each VM, to minimize the risk of Service Level Agreement (SLA) violations and performance degradation due to under/overloaded servers. In addition, multi-objective VM placement and migration algorithms are proposed to reduce the network traffic and power consumption within datacenter. The proposed framework is evaluated by executing experiments on three real world workload datasets namely, Google Cluster Dataset, Planet Lab and Bitsbrain traces. The comparison of proposed framework with the state-of-art approaches, reveals its superiority in terms of different performance metrics. The improvement in power saving achieved by OP-MLB framework is upto 85.3% over the Best-Fit approach.

The Internet of Things (IoT) allows for the coordination and orchestration of a large number of physical and virtual objects connected to the Internet. Sensors used in IoT applications may be in remote locations with an unstable network infrastructure or with limited management of their functions due to processing overload on devices responsible for performing their management (i.e. IoT Gateways) through IoT services. A failure or overload on a network link associated with an IoT Gateway, or failure on the IoT Gateway itself due to processing overload, may result in temporary or even permanent unavailability in IoT applications (Web, Mobile, and Desktop). To address these problems, this paper presents a solution for the need to execute load balancing for the Fog of Things Platforms (in the Fog, and between Fog and Cloud) via the programmability through the use of Software-Defined Networks (SDN). The IoT environment proposed in this work is based on the Fog of Things (FoT) paradigm, which proposes the provision of processing, storage and IoT services at edge devices (local network), optimizing , when possible, the use of IoT services in the Cloud.

Real-time data transmission is one of the objectives of MANET (mobile ad-hoc network) to handle emergencies like a forest fire, flood, and earthquake. In this scenario, quick delivery of data is itself a challenging task for MANET and there is a possibility of load imbalance due to packet transmission in between source-destination pairs via the shortest path due to congestion or control overhead. In this paper, a novel routing protocol, called ACOLBR (ACO based load balancing routing), is design to control the congestion and balancing the load among the multiple paths in between source to destination. The similarity between the environment of Ant Colony and the MANET inspires the authors to apply ACO (Ant colony optimization) technique during routing in MANET to control congestion and balance the load in the network. In our proposal, two colonies of ants (red/blue) carry their packets based on the network condition. A decision variable is designed to select red/blue ant for transferring packets based on different network parameters such as bandwidth, energy, mobility, and distance. The selection of red ant means the route where the concentration of red pheromone is maximum and reverse is true for blue ant. This protocol is also concerns about the link failure during packet transmission in a route. Simulation results using OMNET++ show that ACOLBR outperforms ARA, ANTHOCNET, FACO, AODV, DSDV, DSR, CA-ARTT, and MOAODV in terms of load balancing efficiently in the route for data transmission in between source to destination.

The evolution of wireless body area network (WBAN) has changed the human life for its applications in the field of healthcare, fitness, entertainment and sports etc. However, two of the major challenges in the design of WBAN are energy efficiency and guaranteeing QoS. The load balance of different packet queues in a WBAN plays important role in design of energy efficient and reliable system. In this paper, we propose a Traffic Prioritized Load Balanced Scheduling (TPLBS) algorithm for load balancing in different priority queues in wireless body area networks based on IEEE 802.15.6 model. The main objective of this paper is to minimize packet drop in the queues so that throughput of WBAN can be improved. In this case we consider both priority and location from which the packet received, so that no packet should wait for a long time in the assigned queue to be transmitted to Access Point. The proposed method is simulated using Castalia to study and compare its performance with its counterparts. The simulation results reveal that the proposed protocol works better than the existing methods in terms of delay, throughput and energy efficiency.

Datacenters receive dynamic workloads with disparate specifications to be scheduled on virtual machines (VMs). These unpredictable, alarmingly growing workloads with varying resource specifications may bring down the servers of datacenters into an imbalanced state. Thus, resulting in low resource utilization and high energy consumption among the servers. To cater to the need of fluctuating on-demand resource provisioning, it is essential to scale up the ability and capacityof existing infrastructure through virtualization. Moreover, due to the involvement of conflicting scheduling constraints, load scheduling in cloud computing fall under NP-hard problem. An effective scheduling mechanism in amalgamation with a load balancing strategy based on binary JAYA is implemented to alleviate the challenges as mentioned above. This technique not only improves resource utilization but also brings down the degree of energy consumption and makespan while keeping the whole system balanced. At first, it focuses on evoking a load balancing procedure to uniformly disperse the loads among VMs based on the compatibility between the tasks and VMs and secondly, JAYA algorithm is executed to find the best possible mapping of tasks onto VMs. In order to appraise the efficacy of the proposed algorithm, the scheduling of independent and non-preemptive tasks is simulated in CloudSim using a benchmark parallel workload by NASA-iPSC. Experiments are conducted in both homogenous and heterogeneous environments. The proposed algorithm is statistically validated using the Friedman test, followed by a Holm’s test. The proposed approach is compared over other algorithms such as Round Robin (RR), binary Particle Swarm Optimization (PSO) and Genetic Algorithm (GA). The simulation results show a notable amelioration over other mentioned algorithms by an increase of resource utilization with 18.47% (GA), 12.65% (BPSO) and 4.18% (GA), 2.51% (BPSO) and a reduction of makespan by 6.47% (GA), 4.35% (BPSO) and 4.17% (GA), 2.20% (BPSO) with an increasing number of tasks and VMs in two different test cases respectively.

With the increasing of the scale of task or request and dynamic nature of cloud resources, it gives significant issues of load balancing, resource utilization, task allocation, and system performance and so on. To solve those problems many researchers have applied different types of scheduling techniques. But meta-heuristic scheduling is the most accomplish preferred outcomes over conventional heuristics and hybrid scheduling. Among various meta-heuristics algorithms, PSO is a famous metaheuristic technique to solved optimization issue. PSO is appropriate for dynamic task scheduling, workflow scheduling and load balancing. PSO has a strong worldwide searching capability toward the start of the run and a nearby pursuit close to the furthest limit of the run. Therefore, it has been generally utilized in different applications and has made incredible progress. In this paper a systematically reviews is done on different types of particle swarm optimization (PSO) based scheduling strategy with set of challenges and future direction.

One of the operational challenges in the Internet of Things (IoT) is load balancing, which is the focus of interest of this paper. We propose a three-phase enhanced ad hoc on-demand distance vector (enhanced-AODV) routing protocol for multi wireless sensor networks (multi-WSNs). The three phases are categorized based on traffic priority, namely: high priority, low priority, and ordinary network traffic. The network architecture is divided into chains, i.e. local and public chains, where the cluster heads (CHs) and base stations (BSs) are used respectively to manage network traffic based on priority information with alternative route allocation. Moreover, our three-phase enhanced-AODV protocol provides traffic categorization with alternatives route allocation to minimize energy consumption and prolong the lifetime of participating devices in the network. The proposed model is implemented in the simulation environment to overview results statistics in terms of network lifetime, prioritize traffic, computation and communication costs, latency, and packet lost ratio (PLR). Findings from the simulation suggest that our scheme achieves 15% improvement in network lifetime, 17% latency, 22% PLR, and approximately 10% in the computation and communication costs of the network, in comparison to three other similar protocols.

This article describes the capability of online data storage which has been enhanced by the emergence of cloud datacenter development. Distributed Hash Table (DHT) based image retrieval system using locality sensitive hash (LSH) has provided an efficient way to set up distributed Content Based Image Retrieval (CBIR) frameworks. However, with the fixed LSH function adopted, LSH and other codebook-based distributed retrieval systems are facing the problem of flexibility, and also are difficult to satisfy the user's demand. In this article, LRFMIR is proposed to introduce semantic search into DHT based CBIR system. LRFMIR is established on a DHT based network, where a flexible result truncating strategy is employed to fuse provided results by using multiple features measurements. Experiments show that LRFMIR provides a higher accuracy and recall rate than single feature employed retrieval systems, and possesses good load balancing and query efficiency performance.

In recent years, research has been conducted on dynamic map systems, which are information and communication platforms that manage vehicle sensor information and run applications. However, there is a concern about scalability in the dynamic map system, which operates on a cloud on the Internet, when the number of vehicles that transmit and receive sensor information increases. Therefore, it is considered that the application load can be distributed by distributing multiple edge servers geographically and having the information managed in the cloud by the edge servers. However, the edge server that controls the vehicle information and the edge server that receives the data do not always match depending on the actual radio wave conditions. Therefore, some applications are difficult to manage, such as intersection collision danger warning and merging arbitration when edge servers are assigned in the same way as base stations that receive data from vehicles. Such an application should receive all vehicle data for the target road area. Therefore, we divide the area on the road where a vehicle travels as ``lane section ID'' and assign an edge server to each lane section ID on the basis of that area. In addition, we implemented a dynamic map system that connects vehicles and edge servers by linking multiple edge servers. This enables the edge server to aggregate vehicle data without being affected by radio traffic conditions. We evaluated the scalability of the dynamic map system and verified the effectiveness of the load balancing mechanism using multiple edge servers.

When a set of self-interested users shares multiple resources in a distributed system, we face the problem of allocating resources, called the load balancing problem. In particular, load balancing is defined as allocating the load to the servers of the distributed system such that jobs' response time is minimized, and the utilization of servers is improved. In this paper, the load balancing problem in a distributed system consists of a finite set of servers, and a finite set of users is studied. The load balancing problem considered here is a bi-objective problem with two highly probable conflicting objectives: (i) minimizing jobs' response time (ii) providing the fair utilization of servers. In order to satisfy these two objectives simultaneously, both the objectives are considered in an integrated manner. Next, the load balancing problem is formulated as a noncooperative game; and to solve the game (i.e., to find the Nash equilibrium), a distributed load balancing algorithm (DLBA) is proposed. An experimental study is carried out to ascertain the efficacy of the proposed DLBA. Further, we compare DBLA with three existing load balancing approaches to evaluate its comparative effectiveness. The experimental results validate the effectiveness of the DLBA over the existing approaches.

In cloud computing environment tasks are allocated among virtual machines (VMs) having different length, starting time and execution time. Therefore, balancing these loads among VM is a key factor. Load balancing has to be carried out in such a manner that all VMs should have balanced to achieve optimal utilization of its capabilities and improve the system performance. In this paper, we proposed a load balancing technique by using modified PSO task scheduling (LBMPSO) to schedule tasks over the available cloud resources that minimizes the makespan and maximizes resource utilization. This is achieved by having proper information among the tasks and resources within the datacenter. Our proposed scheduling algorithm is implemented by using CloudSim simulator. Simulation results clearly shows that proposed scheduling algorithm performs better in reducing makespan and increases the resource utilization than other existing techniques.

Cloud Computing is a robust model that allows users and organizations to purchase required services per their needs. The model offers many services such as storage, platforms for deployment, convenient access to web services, and so on. Load Balancing is a common issue in the cloud that makes it hard to maintain the performance of the applications adjacent to the Quality of Service (QoS) measurement and following the Service Level Agreement (SLA) document as required from the cloud providers to enterprises. Cloud providers struggle to distribute equal workload among the servers. An efficient LB technique should optimize and ensure high user satisfaction by utilizing the resources of VMs efficiently. This paper presents a comprehensive review of various Load Balancing techniques in a static, dynamic, and nature-inspired cloud environment to address the Data Center Response Time and overall performance. An analytical review of the algorithms is provided, and a research gap is concluded for the future research perspective in this domain. This research also provides a graphical representation of reviewed algorithms to highlight the operational flow. Additionally, this review presents a fault-tolerant framework and explores the other existing frameworks in the recent literature.

Load balancing is directly associated with the overall performance of a parallel and distributed computing system. Although the relevant problems in communication and computation have been well studied in data center environments, few works have considered the issues in an Internet of Things (IoT) edge scenario. In fact, processing data in a load balancing way for the latter case is more challenging. The main reason is that, unlike a data center, both the data sources and the network infrastructure in an IoT edge system can be dynamic. Moreover, with different performance requirements from IoT networks and edge servers, it will be hard to characterize the performance model and to perform runtime optimization for the whole system. To tackle this problem, in this work, we propose a load-balancing aware networking approach for efficient data processing in IoT edge systems. Specifically, we introduce an IoT network dynamic clustering solution using the emerging deep reinforcement learning (DRL), which can both fulfill the communication balancing requirements from IoT networks and the computation balancing requirements from edge servers. Moreover, we implement our system with a long short term memory (LSTM) based Dueling Double Deep Q-Learning Network (D3QN) model, and our experiments with real-world datasets collected from an autopilot vehicle demonstrate that our proposed method can achieve significant performance improvement compared to benchmark solutions.

Load balancing of tasks on the cloud environment is an important aspect of distributing resources from a data centre. Due to the dynamic computing through the internet; cloud computing agonizes from overloading of requests. Load balancing has to be carried out in such a manner that all virtual machines (VM) should have balanced load to achieve optimal utilization of its capabilities. This paper proposes a novel methodology of dynamic balancing of load among the virtual machines using hybridization of modified Particle swarm optimization (MPSO) and improved Q-learning algorithm named as QMPSO. The hybridization process is carried out to adjust the velocity of the MPSO through the gbest and pbest based on the best action generated through the improved Q-learning. The aim of hybridization is to enhance the performance of the machine by balancing the load among the VMs, maximize the throughput of VMs and maintain the balance between priorities of tasks by optimizing the waiting time of tasks. The robustness of the algorithm has been validated by comparing the results of the QMPSO obtained from the simulation process with the existing load balancing and scheduling algorithm. The comparison of the simulation and real platform result shows our proposed algorithm is outperforming its competitor.

Targeting at implementing the next generation radio access networks (RANs) with virtualized network components, the open RAN (O-RAN) has been regarded as a novel paradigm towards fully open, virtualized and interoperable RANs. Through particularly introducing RAN intelligent controllers (RICs), machine learning (ML) can be unprecedentedly installed, adapting to various vertical applications and deployment environments without sophisticated planning efforts. However, the O-RAN also suffers two critical challenges of load balancing and frequent handovers in the massive base station (BS) deployment. In this paper, an intelligent user access control scheme with deep reinforcement learning (DRL) is proposed. To optimize the performance of distributed deep Q-networks (DQNs) trained by user equipments (UEs), a federated DRL-based scheme is proposed with a global model server installed in the RIC to update the DQN parameters. To further predictively train a global DQN with acceptable signaling overheads, the upper confidence bound (UCB) algorithm to select the optimal UE set and a dueling structure to decompose the DQN parameters are developed. With the proposed scheme, each UE effectively maximizes the long-term throughput and avoids frequent handovers. The simulation results well justify the outstanding performance of the proposed scheme over the-state-of-the-arts, to serve as references for the O-RAN standardization.

The emergence of Network Function Virtualization (NFV) and Service Function Chaining (SFC) together enable flexible and agile network management and traffic engineering. Recently, Network Function Parallelism (NFP) has been proposed to break the need for sequential services and, hence, significantly reduce the latency of SFC. While some studies have investigated how to identify parallel paths to enhance the efficiency of parallelism, less effort has been paid to efficient network function embedding with consideration of parallelism opportunities. Hence, this work aims at solving the instance deployment problem so as to prevent parallel functions from waiting for each other before continuing to the next stage. As it is extremely difficult to optimize VNF (Virtual Network Functions) embedding for uncertain parallelism opportunities, we propose a practical embedding scoring mechanism to enhance the likelihood of low-cost function parallelism. Our scoring design tends to cluster independent functions for efficient parallelism while ensuring load balancing. After instance deployment, we then assign function instances to parallel chains so as to minimize their end-to-end latency while balancing the workload of instances. Our evaluation results show that the proposed scoringbased embedding scheme ensures homogeneous delays of parallel functions and, hence, reduces the end-to-end latency for NFP by up to 22% as compared to the embedding algorithm without considering parallelism opportunities.

Owing to the nonoverlapping spectrum, light fidelity (LiFi) and WiFi technologies can coexist and form a heterogeneous LiFi WiFi network (HLWN). The performance of HLWN significantly depends upon the load balancing strategies. Since load balancing of HLWN is a nonconvex mixed-integer nonlinear programming optimization problem, it is mathematically intractable, and therefore, the conventional optimization methods fail to provide an optimal global solution. Although an optimal solution can be obtained using the exhaustive search method, it would be computationally complex. Therefore, in this article, a reinforcement learning (RL)-based algorithm is explored for solving the load balancing problem for the downlink HLWN at reasonably low complexity and near optimal performance. We have proposed three different reward functions for RL; the first and second reward functions work toward maximizing average network throughput and user satisfaction, respectively. The third reward function is designed to maximize the long-term system throughput and ensure at least 50% user's satisfaction for all users. In order to study the effects of link aggregation on the system performance, this article considers two different types of receiver schemes, namely, single access point (SAP) and link aggregation (LA) scheme. While the SAP allows the user to receive data only from an SAP, the LA scheme allows the user to receive data simultaneously from both LiFi and WiFi AP. This article also includes effect of random orientation of the receiver device and handover overhead. Furthermore, concepts of domain knowledge have been included in this article to reduce the computational complexity of the algorithm. The proposed system performance is compared with the the following two benchmarks: received signal strength (RSS) and exhaustive search based on the computational complexity, average system throughput, and user satisfaction. It is shown that the proposed RL scheme outperforms the RSS scheme in average system throughput and user satisfaction. The RL scheme with an appropriate reward function provides a matching performance to the exhaustive search at reasonably low complexity.

Benefiting from dense connections in data center networks (DCNs), load balancing algorithms are capable of steering traffic into multiple paths for the sake of preventing traffic congestion. However, given each path’s time-varying and asymmetrical traffic state, this may also lead to worse congestion when some paths are overutilised. Especially in the two-tier hybrid optical/electrical DCNs (Hoe-DCNs), the port contentions and large-grained optical packets of the fast optical switch (FOS) require the top-of-rack (TOR) switch to have microsecond-level load balancing capability for microburst traffic. This paper establishes a leaf-spine Hoe-DCN model to illustrate the principal characteristic of dynamic load balancing in TOR switches for the first time. Moreover, we propose the dynamic distributed multi-path (DDMP) load balancing algorithm that relies on dynamic hashing computing for network flow distribution in DCNs, which dynamically adjusts traffic flow distribution at microsecond level according to the inverse ratio of the buffer occupancy. The simulation results show that our proposed algorithm reduces the TOR-to-TOR latency by 15.88% and decreases the packet loss by 22.06% compared to conventional algorithms under regular load conditions, which effectively improves the overall performance of the Hoe-DCNs. Moreover, our proposed algorithm prevents more than 90% packet loss under low load conditions.

The emerging uplink (UL) and downlink (DL) decoupled radio access networks (RAN) has attracted a lot of attention due to the significant gains in network throughput, load balancing and energy consumption, etc. However, due to the diverse vehicular service requirements in different vehicle-to-everything (V2X) applications, how to provide customized cellular V2X services with diversified requirements in the UL/DL decoupled 5G and beyond cellular V2X networks is challenging. To this end, we investigate the feasibility of UL/DL decoupled RAN framework for cellular V2X communications, including the vehicle-to-infrastructure (V2I) communications and relay-assisted cellular vehicle-to-vehicle (RAC-V2V) communications. We propose a two-tier UL/DL decoupled RAN slicing approach. On the first tier, the deep reinforcement learning (DRL) soft actor-critic (SAC) algorithm is leveraged to allocate bandwidth to different base stations. On the second tier, we model the QoS metric of RAC-V2V communications as an absolute-value optimization problem and solve it by the alternative slicing ratio search (ASRS) algorithm with global convergence. The extensive numerical simulations demonstrate that the UL/DL decoupled access can significantly promote load balancing and reduce C-V2X transmit power. Meanwhile, the simulation results show that the proposed solution can significantly improve the network throughput while ensuring the different QoS requirements of cellular V2X.

Current distributed file systems are designed to support PB-scale even EB-scale data storage. Metadata service, which manages file attribute information and the global namespace tree, is crucial to system performance. Distributed metadata management, using multiple metadata servers (MDS's) to store metadata, provides effective approaches to alleviate the workload of a single server. However, maintaining good metadata locality and keeping load balancing among MDS's is a nontrivial problem. In this paper, we present the first machine learning based model called DeepHash, which leverages the neural network to learn a locality preserving hashing (LPH) mapping scheme. DeepHash first converts the metadata nodes to feature vectors by the network embedding technology. Due to the absence of training labels, i.e., the hash values of metadata nodes, we design a pair loss function with distinctive characters to train DeepHash, and introduce the sampling strategy to improve the training efficiency. Besides, we propose an efficient algorithm to dynamically balance the workload and adopt the cache model to improve query efficiency. The experiments on the Amazon EC2 platform demonstrate that the DeepHash can preserve the metadata locality meanwhile maintaining a high load balancing, which denotes the effectiveness and efficiency of DeepHash compared with traditional and state-of-the-art schemes.

Large-scale machine learning and data mining applications require computer systems to perform massive matrix-vector and matrix-matrix multiplication operations that need to be parallelized across multiple nodes. The presence of straggling nodes---computing nodes that unpredictably slow down or fail---is a major bottleneck in such distributed computations. Ideal load balancing strategies that dynamically allocate more tasks to faster nodes require knowledge or monitoring of node speeds as well as the ability to quickly move data. Recently proposed fixed-rate erasure coding strategies can handle unpredictable node slowdown, but they ignore partial work done by straggling nodes, thus resulting in a lot of redundant computation. We propose a rateless fountain coding strategy that achieves the best of both worlds---we prove that its latency is asymptotically equal to ideal load balancing, and it performs asymptotically zero redundant computations. Our idea is to create linear combinations of the m rows of the matrix and assign these encoded rows to different worker nodes. The original matrix-vector product can be decoded as soon as slightly more than m row-vector products are collectively finished by the nodes. Evaluation on parallel and distributed computing yields as much as three times speedup over uncoded schemes.

This work studies the quality of service (QoS) performance of wireless networks that integrate light fidelity (LiFi) and wireless fidelity (WiFi). While the hybrid network is potential for improving network capacity, load balancing becomes essential and challenging due to the nature of heterogeneous access points (APs). A number of studies have been conducted to address this issue, focusing on maximising the network capacity with user fairness constraints. However, in practice, QoS metrics including packet loss ratio and latency are important to network services. In this paper, QoS-driven load balancing is studied for hybrid LiFi and WiFi networks (HLWNets) in two scenarios: single-AP association (SA) and multi-AP association (MA). In each case, an optimisation problem is formulated to minimise the packet loss ratio and latency, and a low-complexity iterative algorithm is proposed to solve the problem. Results show that the novel methods, especially MA, can effectively balance the traffic loads among the APs and improve the QoS performance. In addition, the more subflows the better performance MA provides. Targeting the same level of QoS, MA can achieve a system throughput up to 160% higher than the signal strength strategy and 130% higher than the proportional fairness load balancing.

The rapid development of the Industrial Internet of Things (IIoT) enables IIoT devices to offload their computation-intensive tasks to nearby edges via wireless Base Stations (BSs) and thus relieve their resource constraints. To better guarantee Quality-of-Service (QoS), it has become necessary to cooperate multiple edges instead of letting them work alone. However, the existing solutions commonly use a centralized decision-making manner and cannot effectively achieve good load balancing among massive edges that are widely distributed in IIoT environments. This results in long decision-making time and high communication costs. To address this important problem, we propose a Reinforcement learning empowered Feedback control method for Cooperative Load Balancing (RF-CLB). First, by integrating Reinforcement Learning (RL) and Machine Learning (ML) algorithms, each edge independently schedules tasks and performs load balancing between adjacent edges based on the local information. Next, through feedback control and multi-edge cooperation, the objective multi-edge load balancing plan for IIoT can be found. Simulation results demonstrate that the RF-CLB chooses the adjustment operations of load balancing with 96.3% correctness. Moreover, the RF-CLB achieves the near-optimal performance, which outperforms the classic ML-based and rule-based methods by 6~9% and 10~12%, respectively.

Live migration of Virtual machines (VMs) has become a regular tool for edge and cloud operators to facilitate system maintenance, fault tolerance, and load balancing, with little impact on running instances. However, the potential security risks of live migration of VMs are still obscure. In this paper, we expose a new vulnerability in the existing VM live migration approaches, especially the post-copy approach. The entire live migration mechanism relies upon reliable TCP connectivity for the transfer of the VM state. We demonstrate that, if the host server is vulnerable to off-path TCP attacks, the loss of TCP reliability leads to VM live migration failure. We demonstrate that, by intentionally aborting the TCP connection, attackers can cause unrecoverable memory inconsistency for post-copy, leading to a significant increase in downtime and performance degradation of the running VM. Additionally, we present detailed techniques to reset the migration connection under heavy networking traffic. We also propose effective defenses to secure the VM live migration. Our experimental results demonstrate that memory inconsistencies could be devastating to some applications, and it only takes a few minutes to reset a heavy migration connection.

Because of the limited memory of the increasing amount of information in current wearable devices, the processing capacity of the servers in the storage system can not keep up with the speed of information growth, resulting in low load balancing, long load balancing time and data processing delay. Therefore, a data load balancing technology is applied to the massive storage systems of wearable devices in this paper. We first analyze the object-oriented load balancing method, and formally describe the dynamic load balancing issues, taking the load balancing as a mapping problem. Then, the task of assigning each data node and the request of the corresponding data node’s actual processing capacity are completed. Different data is allocated to the corresponding data storage node to complete the calculation of the comprehensive weight of the data storage node. According to the load information of each data storage node collected by the scheduler in the storage system, the load weight of the current data storage node is calculated and distributed. The data load balancing of the massive storage system for wearable devices is realized. The experimental results show that the average time of load balancing using this method is 1.75 ​h, which is much lower than the traditional methods. The results show the data load balancing technology of the massive storage system of wearable devices has the advantages of short data load balancing time, high load balancing, strong data processing capability, short processing time and obvious application.

Program execution on a multicore platform, even a homogeneous one, does not guarantee performance gains. The key is in letting each of the cores contribute to the overall computation for the maximum amount of time. For anything but trivial problem settings, and especially for heterogeneous platforms like ones made of CPU and GPU cores, this calls for some form of load balancing, i.e., explicit or implicit workload shifts between the cores, so that idle times are eliminated to the maximum degree possible. In this chapter we examine the two generic categories of load balancing algorithms: static which can be considered pro-active, meaning that they pre-calculate load assignments without expensive communications taking place, and dynamic which can be considered re-active as they can adapt to changes in the execution environment. The chapter is concluded by two cases studies that show how both static and dynamic load balancing can be used to evenly distribute the work between the nodes of a heterogeneous platform, that combines both CPU and GPU nodes.

Job dispatching is a fundamental problem in edge computing for load balancing among multiple edge servers. When implementing an edge computing system with distributed job dispatchers in a sizable network, such as a Metropolitan Area Network (MAN), the highly dynamic transmission latency is non-negligible, which could lead to outdated information being shared. Moreover, the fully-observed system state is beyond reach as the reception of any broadcast is time consuming. In this paper, we investigate the online distributed job dispatching problem in edge computing, where multiple access points (APs) collect jobs and then dispatch each job to an edge server. The distributed dispatcher on each AP would receive partially and outdated information exchanged via periodic broadcast. Hence, we formulate the distributed job dispatching problem by leveraging the partially observable Markov decision process (POMDP) and propose a novel approximate MDP solution framework, called DecMDP, that bypasses the huge time complexity of conventional POMDP solutions. Both analytical and semi-analytical performance lower bounds are derived for the approximate MDP solution. Furthermore, we extend DecMDP to handle a more general scenario where a priori knowledge of the system is absent. Finally, extensive simulations based on the Google Cluster traces show that our policy can achieve the best performance when compared with heuristic baselines, e.g., achieving 20.67% reduction in average job response time, and consistently performs well under various parameter settings.

The network function virtualization paradigm replaces dedicated hardware appliances with software running on virtual machines in commodity servers. However, to overcome performance and scalability issues on virtual network functions (VNFs) demanding intensive packet processing, we propose implementing these VNFs as programmable rules distributed among software-defined networking (SDN) switches. Our approach allows infrastructure providers to combine hardware-based and software-based SDN switches, exploring the tradeoff between the packet processing speed and instantiation flexibility. Our proposal includes a scalable mechanism, which decides the type and number of switches to meet network demand, along with a load balancing mechanism to distribute flows among switches, considering their singularities. We first evaluate our proposal in a small testbed with real equipment, confirming its effectiveness. Then, we evaluate our proposal with simulations to further investigate its scalability. Our results show that, even under high traffic loads, the joint operation of both mechanisms improves VNF capacity and saves virtualization resources.

The Software Defined Networking (SDN) paradigm decouples the network control functions from the data plane and offers a set of software components for flexible and controlled management of networks. SDN has promised to provide numerous benefits in terms of on-demand provisioning, automated load balancing, streamlining physical infrastructure, and flexibility in scaling network resources. In order to realize these network service offerings, there is an important need for developing an efficient, robust, and secure execution platform. As a primary contribution, we present a novel virtual execution platform for the OpenFlow controller using Network Function Virtualization (NFV). Theoretically, NFV can apply to any network function, which can simplify the managing of the heterogeneous data plane. The characteristics of our proposed architecture include pipe-lined processing of network traffic, virtualized and replicated execution of network functions, isolation between task nodes, and random mapping of traffic to task nodes. The proposed architecture has two major components: a Network Packet Schedulers (NPS) and a Task Engine (TE). The TE consists of Task Nodes (TNs) which are responsible for executing different network functions on various traffic flows and each TN is realized as a virtual machine. Upon receiving traffic from the data plane, NPS analyses the functional requirements of the traffic and different controller performance parameters. Then it allocates the traffic to appropriate TNs for executing necessary network functions. In this respect, it provides performance benefits, robustness, fine-grained modularity, and strong isolation security in the processing of traffic flows on the SDN platform. Efficacy of our proposed architecture has been demonstrated with a case study.

Insecure and portable devices in the smart city’s Internet of Things (IoT) network are increasing at an incredible rate. Various distributed and centralized platforms against cyber-attacks have been implemented in recent years, but these platforms are inefficient due to their constrained levels of storage, high energy consumption, the central point of failure, underutilized resources, high latency, etc. In addition, the current architecture confronts the problems of scalability, flexibility, complexity, monitoring, managing & collecting of IoT data and defend against cyber-threats. To address these issues, the authors present a distributed and decentralized Blockchain-Software Defined Networking (SDN) based energy-aware architecture for IoT in smart cities. Thus, SDN continuous observing, controlling, managing IoT devices activities and detect possible attacks in the network; Blockchain provides adequate security & privacy against cyber-attacks, reduces the central point of failure issues; Network Function Virtualization (NFV) are used to saving energy, load balancing, as well as increasing the lifetime of the entire network. Also, we introduce a Cluster Head Selection (CHS) algorithm to reduce the energy consumption in the presented model. Finally, we analyze the performance using various parameters (e.g., throughput, response time, gas consumption, communication overhead) and demonstrating the result that provides higher throughput, lower response time, lower gas consumption than existing works for smart cities.

The further evolution and integration of internet of things (IoT) and artificial intelligence (AI) technologies would transform traditional health care processes into smart health care services, which leverages various wireless sensors, cameras, and controllers to collect multi-dimensional medical information and analyze them intelligently. In case of this, the massive amount of data generated by smart health devices need to be transferred, processed and stored quickly, which poses great challenges to traditional cloud-based communications with single association mechanism. To address this problem, this paper proposes an innovative multi-access edge computing based access network with multi-association mechanism, where a single smart device could associate and communicate with multiple edge nodes. The deployment of edge nodes for proposed access network is formulated as an extended knapsack problem, which considers not only signal to interference plus noise rate (SINR) requirement of devices, but also power constraint and load balance of edge nodes and it is solved by the simulated annealing (SA) algorithm. Numerical results via SA algorithm show that the proposed multi-association mechanism can improve the SINR of devices significantly, while achieving better load balancing among edge nodes. All these together provides good reference to build an efficient access network for future smart healthcare system.

In this contemporary era, recent developments have seen proliferation of methods on improving energy utilization for wireless sensor networks. Although the proposed methods can mitigate the problem of rapid battery depletion to some extent, the lifespan of sensor nodes is still a biggest constraint. Many studies conducted so far have shown continuous network functions with the help of external harvesting techniques. However, these provide low output because of the limitation on the energy capturing devices. Therefore, energy limitation strongly restricts the usage of wireless sensor nodes. This paper aims to provide high recharging rates and better energy efficiency by proposing a three-step mechanism, which is an extension of our existing proposed work J-ERLB (Joint Energy Replenishment and Load Balancing). In a three-step mechanism, topology selection followed by recharging and load balancing are combined together to prolong the lifetime of sensor nodes. Topology selection is performed by implementing J-ERLB on various topologies like ring, star, and cluster. Further, taking advantage of Unmanned Aerial vehicle (UAV) we focus on data collection and high-rate recharging. We have implemented System on Chip (SoC) integrated chip on UAV to achieve goal of perpetual network operations. Finally, the effectiveness of proposed mechanism is tested by comparing its numerical outcomes with existing J-ERLB and J-MERDG (Joint Mobile Energy Replenishment and Data Gathering) techniques. The overall throughput for 3SM is doubled in comparison with the existing techniques, whereas the average remaining energy shows 6 times better savings for the proposed solution in comparison to the existing J-ERLB and J-MERDG when varying data rates is applied.

In wireless sensor networks (WSNs), topology control is of great significance for energy efficiency. While various clustering algorithms have been proposed to handle the topology control of WSNs, the joint optimization of fairness, quality of service (QoS), load balancing and congestion management has received much less attention. Additionally, the constrained resources and processing capacity of sensor nodes complicate the topology control problem. This paper proposes a three-layer framework based on joint rate-aware fuzzy clustering and stable sensor association that considers various factors of sensor energy efficiency. First, a rate-aware fuzzy clustering method is proposed for initial clustering, and energy-aware cluster head selection is applied considering the total energy consumption and the residual energy of the potential cluster heads to improve the energy efficiency of WSNs. After that, the payoff functions of the cluster heads and the cluster members are constructed for sensor association with resource capacity limitations and QoS constraints. The optimal sensor association is obtained to maximize the payoffs. Furthermore, a low-complexity suboptimal sensor association approach is proposed to reduce the complexity with a tolerable performance gap. Finally, a congestion factor is introduced to balance the load of cluster heads, and the Gale-Shapley (GS) algorithm is used to avoid unstable sensor association caused by the processing capacity constraints. The simulation results show that the proposed algorithms can effectively improve the resource utilization and the energy efficiency of the WSNs.

Load balancing among virtual machines (VMs) is significant for delivering the cloud services in optimized way with minimum cost paid and total time acquired to deliver the services. In this paper, the various research gaps for load balancing optimization in the past literature have been presented, which need to be addressed for solving the load balancing problem in cloud environment. In present work, Hybrid approach based resource provisioning and load balancing framework for workflows execution has been proposed to optimize the utilization of VMs with uniform load distribution. The proposed framework is based on the hybridization of heuristic techniques with metaheuristic algorithm to achieve its optimal performance in terms of makespan and cost. Two hybrid approaches have been proposed for HDD-PLB framework-Hybrid Predict Earliest Finish Time (PEFT) Heuristic with Ant Colony Optimization (ACO) metaheuristic (HPA) and Hybrid Heterogeneous Earliest Finish Time (HEFT) heuristic with ACO (HHA). The two proposed approaches for load balancing have been analyzed and compared to determine which is superior for proposed HDD-PLB framework.

This paper considers the steady-state performance of load balancing algorithms in a many-server system with distributed queues. The system has N servers, and each server maintains a local queue with buffer size b-1, i.e. a server can hold at most one job in service and b-1 jobs in the queue. Jobs in the same queue are served according to the first-in-first-out (FIFO) order. The system is operated in a heavy-traffic regime such that the workload per server is $λ = 1 - N^{-α}$ for 0.5≤ α

At present, the amount of data from users is increasing exponentially, and most of the data is stored in data centers distributed in different geographic locations. The cost of transferring large amounts of data across geographically distributed data centers can become prohibitive. Therefore, to shorten the data transmission time and reduce the cost of data transmission bandwidth and maintain the load balance of the geographically distributed cloud system, an optimal data placement strategy considering capacity limitation and load balancing in a geographically distributed cloud is proposed. Firstly, the capacity limitation, load balancing, and bandwidth cost of each cloud data center in the geographically distributed cloud are considered, and the data placement problem in the geographically distributed cloud is mathematically modeled. Secondly, the Floyd algorithm is used to model the cost of data transmission bandwidth and find the minimum transmission bandwidth cost. Finally, the Lagrangian relaxation method is used to obtain the optimal data placement scheme for the transmission time. To show the performance advantages of the proposed algorithm, comparative experiments are carried out. When the bandwidth is 15 Mbps, in terms of the Load Balancing Degree ( L B D ), the proposed algorithm is 40.3% higher than the Hash on average, 35.6% higher than the Closest on average, and 25.7% higher than the CRANE on average. Moreover, the experimental results show that the proposed algorithm can reduce the data transmission cost, and improve the load balancing in a geographically distributed cloud system.

The topologies of current data center networks are typically multi-rooted trees (e.g. leaf–spine) with rich parallel paths between any pair of hosts. Recent progress has demonstrated that effective load balancing can fully utilize these parallel paths to speed up data transfer. Nonetheless, the existing load balancing designs are agnostic to the heterogeneous datacenter traffic, i.e., a mass of delay-sensitive short flows mix with a handful of throughput-oriented long flows, and casually reroute these flows regardless of path condition, thus resulting in frequent flow collisions. The short flows suffer from the problem of large queuing delay due to colliding with long flows, while the frequent collisions between long flows lead to low network utilization and serious throughput degradation. To address these inefficiencies, we propose an isolation-based load balancing scheme, namely ILB, which perceives flow collisions and dynamically isolates the long flows from short flows. Specifically, when a long flow collides with short flows in the same path, it immediately switches to another path unused by short flows to help the short ones in the previous path complete quickly. When short flows disappear, the long flow quickly occupies all its available paths to achieve high throughput. Moreover, ILB is only deployed on the leaf switch without modifying the end-host and spine switch. Experimental results of NS3 simulations show that ILB respectively reduces the average and tail flow completion time for short flows by up to 30% and 70%, as well as increases the throughput of long flows by about 1.68x over the state-of-the-art datacenter load balancing schemes.

The past few years have witnessed a surge of interest towards deep reinforcement learning (DRL) in computer networks. With extraordinary ability of feature extraction, DRL has the potential to re-engineer the fundamental resource allocation problems in networking without relying on pre-programmed models or assumptions about dynamic environments. However, such black-box systems suffer from poor robustness, showing high performance variance and poor tail performance. In this work, we propose a unified Teacher-Student learning framework that harnesses rich domain knowledge to improve robustness. The domain-specific algorithms, less performant but more trustable than DRL, play the role of teachers providing advice at critical states; the student neural network is steered to maximize the expected reward as usual and mimic the teacher’s advice meanwhile. The Teacher-Student method comprises of three modules where the confidence check module locates wrong decisions and risky decisions, the reward shaping module designs a new updating function to stimulate the learning of student network, and the prioritized experience replay module to effectively utilize the advised actions. We further implement our Teacher-Student framework in existing video streaming (Pensieve), load balancing (DeepLB), and TCP congestion control (Aurora). Experimental results manifest that the proposed approach reduces the performance standard deviation of DeepLB by 37%; it improves the 90th, 95th, and 99th tail performance of Pensieve by 7.6%, 8.8%, and 10.7% respectively; and it accelerates the growth rate of Aurora by 2x at the initial stage, and achieves a more stable performance in dynamic environments.

This paper proposes a new cross-layer communication system for the provision of Internet services and applications to mitigate the negative impacts of COVID-19, due to which the massive online demands are affecting the current communication systems' infrastructures and capabilities. The system requirements and model are investigated where it utilizes high-altitude platform (HAP) for fast and efficient connectivity provision to bridge the communication infrastructure gap in the current pandemic. The HAP is linked to the main server or gateway station located on ground and can provide communication narrow beams towards isolated areas which suffer from poor terrestrial radio coverage or lack of communication infrastructure. The vital e-learning applications using Internet services provision from the proposed HAP system are described and modelled including system adaptation parameters such as the application and physical layers to control the data rates of different e-learning applications and the overall cell data rate. On the other hand, the provision of high-speed Internet services from the proposed system is supported by using adaptive antenna arrays onboard HAP which provides high-gain beams to achieve the required high-quality transmission data rates at the student premises and provides the capability of coverage cell area adaptation for load balancing. The concentric circular antenna arrays with tapered feeding are proposed in this adaptive antenna system to control the cell mainlobe gain and reduce the out-of-coverage radiation as well. In addition, the system feasibility has been proved in two coverage scenarios including single-beam and multibeam HAP communications.

In this paper, we study big data-driven Cyber–Physical Systems (CPS) through cloud platforms and design scheduling optimization algorithms to improve the efficiency of the system. A task scheduling scheme for large-scale factory access under cloud–edge collaborative computing architecture is proposed. The method firstly merges the directed acyclic graphs on cloud-side servers and edge-side servers; secondly, divide the tasks using a critical path-based partitioning strategy to effectively improve the allocation accuracy; then achieves load balancing through reasonable processor allocation, and finally compares and analyses the proposed task scheduling algorithm through simulation experiments. The experimental system is thoroughly analysed, hierarchically designed, and modelled, simulated, and the experimental data analysed and compared with related methods. The experimental results prove the effectiveness and correctness of the worst-case execution time analysis method and the idea of big data-driven CPS proposed in this paper and show that big data knowledge can help improve the accuracy of worst-case execution time analysis. This paper implements a big data-driven scheduling optimization algorithm for Cyber–Physical Systems based on a cloud platform, which improves the accuracy and efficiency of the algorithm by about 15% compared to other related studies.