How to prolong the network lifetime is an important factor when designing a routing protocol in wireless sensor network.To solve this problem,a novel clustering algorithm based on the improved particle swarm optimization was presented.The algorithm modifies the inertial weight to avoid particles trapping in local optimum.It also takes into account both energy balance and transmission distance,and cooperates relays nodes with cluster heads to reduce the excessive energy consumption of cluster heads.This paper compared the proposed algorithm with other algorithms in various scenarios.Simulation results show that the proposed algorithm has good capability on distributing nodes and balancing cluster system.

Billions of Wi-Fi assess points (APs) have been deployed to provide wireless connection to people with different kinds of mobile devices.Toaccelerate the speed of Wi-Fi connection,mobile devices will send probe requests to discover nearby Wi-Fi APs,and maintain previously connected network lists (PNLs) of APs.Previous studies show that the Wi-Fi fingerprints that consist of probed SSIDs individually will leak private information of users.This paper investigated the privacy caused by the Wi-Fi fingerprints in the wild,and provided a data-driven solution to protect privacy.First,measurement studies were carried out based on 27 million users associating with 4 million Wi-Fi APs in 4 cities,and it was revealed that Wi-Fi fingerprints can be used to identify users in a wide range of Wi-Fi scenarios.Based on semantic mining and analysis of SSIDs in Wi-Fi fingerprints,this paper further inferred demographic information of identified users (e.g.,people’s jobs),telling “who they are”.Second,this paper proposed a collaborative filtering (CF) based heuristic protection method,which can “blur” an user’s PNL by adding faked SSIDs,such that nearby users’ PNLs and Wi-Fi fingerprints are similar to each other.Finally,the effectiveness of the design was verified by using real-world Wi-Fi connection traces.The experiments show that the refined PNLs protect users’ privacy while still provide fast Wi-Fi reconnection.

A cluster-based resource allocation mechanism was proposed for the problem of the limited battery life of the user equipment.The mechanism includs three parts:D2D users are put in disjoint clusters by graph theory,and then the auction algorithm is used to allocate the channel for D2D clusters according to the clustering results.In the end,the power allocation is performed by using non-cooperative game theory model.Simulation results show that the proposed mechanism effectively extends battery life,improves the energy efficiency of the users,and satisfies the requirements of the user transmission rate,thus ensuring the users to obtain higher QoS.

With the broadband trend in wireless communication system,traditional data encryption methods have high computational complexity,and haven’t taken the security of modulation modes of the physical layer into consideration.Aiming at this problem,a modulation protection algorithm based on wireless OFDM systems was proposed from the perspective of physical layer encryption.Modulation protection effect of the proposed algorithm was analyzed respectively in the cases of single-carrier and multi-carrier,and typical modulation identification methods in the cognitive radio were adopted in the simulation to analyze and compare the recognition rate before and after encryption.Theoretical ana-lysis and simulation results show that this algorithm doesn’t change the performance of original system,and has high capacity of modulation protection.

The main feature of sparse code multiple access,i.e.,non-orthogonal multiple access,is supported by overloaded connection with limited resources,which can greatly improve the spectrum utilization.Thanks to the sparsity of the SCMA codebook sets,MPA becomes a basic receiver decoding algorithm.Although there exists a similar bit error ratio (BER) performance between the maximum likelihood (ML) detection scheme and traditional MAP method,the complexity of the exponential calculation is still high.To further reduce the complexity problem,a novel low-complexity detection algorithm based on dynamic edge selection strategy was proposed to reduce unnecessary node operation.In each iteration,the belief propagation stability information of the function node to the variable node in the factor graph model is used to dynamically determine whether a node update operation is required.The simulation results show that the complexity of the dynamic edge selection algorithm is significantly reduced,and the BER can be well balanced.

Aiming at the problem of low data delivery ratio in opportunistic networks,this paper defined and computed the node forwarding utility according to the historical contacts between nodes,e.g.,the number of nodes contacts,the length of contacts and the stability of nodes relations.First,the node with the largest utility value falling into the communication range is selected as the initial forwarding node.Then,other forwarding nodes are selected by their cosine similarities,such that the forwarding nodes can be evenly distributed approximately.On this basis,an opportunistic network forwarding algorithm based on node cosine similarity (ONNCS) was proposed.ONNCS enables the selected forwarding nodes to be evenly distributed and improves the data delivery ratio.The simulation results show that ONNCS has a hig-her forwarding success rate and lower forwarding energy consumption,and the success rate of the algorithm is 5% to 8% higher than other algorithms in this paper.

In order to solve the energy hole problem caused by the unevenness of the link communication load in WSNs routing protocol,a dynamic clustering algorithm based on virtual grid (IDCS) and a dynamic load balancing routing algorithm (DCDLB) for maximizing network life cycle considering data forwarding delay were presented.In IDCS algorithm,the area is divided into several virtual grids according to node communication radius,and the nodes in the same grid form a cluster.The cluster head is chosen by distributed cluster head selection strategy considering node’s energy and location factors,and a dynamic cluster head rotation mechanism based on cluster head’s energy level is introduced for balancing consumption.In DCDLB routing algorithm,the network lifetime is maximized by considering energy consumption balance among cluster heads and multihop data forwarding delay.The simulation results show that DCDLB routing algorithm is superior to LEACH,HEED and CRVB routing algorithms in terms of extending network lifetime and decreasing data forwarding delay.

Filtered orthogonal frequency-division multiplexing (F-OFDM) is a new technology for next generation mobile communication system.F-OFDM can maintain the advantages such as strong anti-interference ability in OFDM system,and is adaptive for different flexible service configuration in the future.However,it’s more sensitive to phase noise than OFDM,which will cause sub-band common phase error (SCPE) and sub-band inter-carrier interference (SICI),thus seriously decreasing the performance of the system.This paper proposed a self-adaptive EM phase noise suppression algorithm (AEM-PNS) to reduce the phase noise in F-OFDM.It includes two sub-algorithms:EM-SCPE and EM-SICI.AEM-PNS can choose suitable phase noise suppression sub-algorithm automatically according to the inserted phase noise instruction symbol (PNIS) and pilot instruction symbol (PIS) in symbol frame.Simulation results show that the proposed algorithm can track phase noise adaptively,reduce the influence caused by phase noise effectively,and keep low complexity and high spectral efficiency simultaneously.

Aiming at the description of IoT services,the existing standard is WSDL(Web Services Description Language) based on the XML(Extensible Markup Language).However,WSDL lacks semantic information and can not describe the IoT services semantically,thus having an impact on the accuracy of service discovery.Among the current exi-sting semantic service description languages,OWL-S (Ontology Web Language for Services) has a most profound and lasting influence.In order to describe services semantically,this paper presented a conversion method from WSDL to OWL-S.This method can convert the existing WSDL file to OWL-S file through operation mapping and ontology mapping.Some test sets and examples verify the effectiveness of the proposed method for converting files.And the precision ratio and recall ratio of the conversion results are better than that of MWSAF method.

Opportunistic network which employs store-carry-and-forward pattern doesn’t consider encounter possibility of destination nodes and passing nodes during message transmission,leading to a large deviation for the estimation on message transmission status.This paper proposed an encounter interval based buffer management strategy.It facilitates the characteristic of exponential distribution for the encounter intervals between nodes,and the number of message co-pies is also considered.So the average delivery probability is estimated.All messages stored in the buffer are sorted according to the average delivery probability when the buffer is overflow,and the messages with low delivery probability will be dropped first for buffer management intent.Simulation results show that the proposed strategy achieves better performance compared to existing schemes in respect of the delivery ratio,average latency and overhead ratio.

Current protocol format inferring methods based on network traffic can only extract flat sequence of keywords,and they do not consider the structural features of message keywords,such as sequential,hierarchical and parallel relation between the keywords.Additionally,the noise in message samples always lead to low recognition accuracy of keywords.This paper presented a method to automatically identify keywords of unknown protocol message and infer the message structure.Based on the collected communication messages of the unknown protocol,the method implements two-phase closed sequential patterns to identify protocol keywords and generate keywords sequence with keyword composition relation,extract sequential,hierarchical and parallel relation of the keywords,and then infer messages structure inference.To ensure recognition accuracy of the keywords,the method analyzes message samples directly containing noise by setting minimum support in keywordsidentification procedure.Experimental results show that the proposed method performs well in keywords identification and message structure inference for both text protocol and binary protocol.

Route availability refers to the probability that a user can get the requested service.With the development of the Internet,a large number of real-time services have emerged,and the requirements for the timeliness of the network is becoming higher and higher,and high demands have been placed on the “self-repairing ability” of the Internet.However,network faults occur frequently,and routing loops and long convergence times may occur during the process of repairing network failures.The repair time is usually between several seconds and tens of seconds,which cannot meet the real-time requirements of the Internet.Therefore,improving routing availability has become an urgent problem to be solved.This paper summarized and analyzed the existing schemes to improve routing availability,and divided these schemes into two major categories,namely passive protection scheme and route protection scheme.The research results at home and abroad were introduced in detail,the advantages and disadvantages of each program were compared,the main contributions and shortcomings of these programs were summarized and analyzed,and the research directions were proposed for further research.

In view of the cave environment,this paper presented a geometric model based on the new environment in the cave.The channel on both sides of the cave gradually widen (narrow) from the entrance to the depth.According to the geometric model,and by means of ray theory,this paper assumed that both sides of the channel surface are approximate smooth,and proposed a random channel model of single transmission and single reception for acoustic signal communication system in cave environment.According to the geometric model,the influence of channel opening angle on channel distribution,instant channel capacity,time autocorrelation function,frequency correlation function,Doppler power spectral density and power delay distribution is studied.The theoretical and simulation results show that compared with the case where both sides of the channel are parallel (i.e.the opening angle of both sides is 0),the statistical characteristics of the acoustic channel wireless communication system will be significantly affected by only a small change in the ope-ning angle of both sides of the channel,and the parallel is a special case of this research content.

In the digital communication system,when a discrete signal uniformly distributed on the constellation is transmitted through the AWGN channel with limited power and noise power spectral density,the maximum information rate can’t reach the Gaussian channel capacity.Aiming for improving transmission rate to approach the channel capacity,a non-uniform distribution design of signal constellation is quite necessary.Therefore,Box-Muller transform is used to construct the M-APSK signal constellation obeying the Gaussian distribution when the constellation points approach infinity .Simulation results show that the maximum information rate is equal to the Gaussian channel capacity.Compared with rectangular M-QAM signal,the performance improves considerably when the modulation order is sufficiently high.On this basis,according to the distribution characteristics of the constellation,this paper designed a modulation and demodulation scheme based on Gray coding and simplified Max-Log LLR algorithm which reduces system complexity significantly.At last,the system complexity and bit error rate performance of the proposed scheme were verified through Matlab simulation.

Improvement of technology makes smart mobile devices more and more popular.Mobile device traffic is growing rapidly.However,due to the limited resources and computing performance of smart mobile devices,mobile device may face the situation of insufficient capacity when dealing with compute-intensive and time-sensitive applications.Unloading the computations which the mobile terminal needs to process to the computing nodes in the edge network for calculation is an effective way to solve this problem.This paper first introduced the existing calculation offloading strate-gies and elaborated from the aspects of minimizing delay,minimizing energy consumption and maximizing benefits.Then,it compared the advantages and disadvantages of different offloading strategies.At last,it considered and prospected the future development of calculation offloading strategies of mobile edge network.

Wireless sensor networks (WSN) are a self-organizing network composed of a large number of sensor nodes with limited communication,computing and storage capabilities,which can be deployed to perform long-term monitoring tasks in harsh environments.Data query processing is one of the most basic operations for WSN to obtain monitoring data.It mainly refers to the user distributing query requests to the network through a specific node,and then the node that satisfies the requirements returns the data to the user.In the process of query,because of the dynamic nature of the network (e.g.,the destruction of nodes by external forces,the movement or sleep,resulting in changes in network topology and connectivity,and unreliable communication links,etc.),the data transmission has large delay,high energy consumption and even data loss,resulting in low success rate of query.At present,many scholars study this problem and make some progress,but there are still many problems to be solved in practical application.In order to further promote the in-depth study of data query technology in dynamic wireless sensor networks,this paper analysed and summarized the typical work in recent years,and compared their advantages and disadvantages.Then,this paper discussed the key issues that need to be solved in this field,and finally pointed out the next research directions.

With the expansion of the Internet industry,research and innovation of network core technologies are urgent.The Future Network Experimentation Facility (FNEF) was designed to provide efficient and convenient resources for network-related researchers to support innovative research and experiments on network technologies.The resources scheduling management system is a very important task in FNEF.Based on the resource scheduling and test service requirement of FNEF,this paper designed a combination of centralized and distributed architecture.The central resource scheduling management system cooperates with the node resource scheduling management system to schedule the backbone network resources and computing resources in each disperse site.This paper proposed a multi-objective optimized resource scheduling algorithm by considering the communication cost between the virtual machines,the physical machine resource utilization and the resource balance.Simulation experiments show that the proposed algorithm can effectively optimize the above multiple objectives.

As vehicle applications,mobile devices and the Internet of Things (IoT) have been developing rapidly,building an efficient architecture to deal with big data in vehicular networks has become an important concern for the future smart city.Howe-ver,the complex and inflexible architecture of vehicular networks faces a set of challenges such as high mobility,intermittent connectivity,heterogeneity of applications.In this context,software defined network (SDN),with the programmable and flexible network architecture,has recently been gaining great attentions from research communities,businesses,and industries in wired network managements and heterogeneous wireless communications.Applying SDN to Vehicular Networks can significantly improve its flexibility,reliability,programmability and scalability,enhance the capacity of vehicular Networks in providing applications and services,and further improve the quality of experience of users.Firstly,the SDN framework was described.Secondly,the research progress of the software defined vehicular network (SDVN) was summarized from two perspectives:architectures and data dissemination.Then the current research state of SDVN combined with mobile edge computing (MEC) was surveyed.After that,exi-sting problems and challenges faced by SDVN were discussed.Finally,several SDVN application prospects were introduced.

In recent years,with the rapid development of modern network communication and social media technologies,complex network has become one of the frontier hotspots of interdisciplinary research.As an important problem in the research of complex network,community detection has important theoretical significance and application value,and has attracted increasing attention.Many community detection algorithms and reviews have been proposed.However,most of the existing reviews on community detection focus on a particular direction or field.On the basis of previous work,this paper did deep research in the community detection algorithms,and gave a review on the research progress of community detection.Firstly,this paper gave the definition of community detection and evaluation measurements for different network structure.Then,this paper introduced the classic community detection algorithms on different network structure,including the global community detection and local community detection algorithms on homogeneous networks,community detection on heterogeneous network,and community detection on link structure combined with node content,as well as the dynamic network community detection and community evolution.Finally,this paper briefly introduced the typical applications of community discovery in the real world,including impact maximization,link prediction and emotion analysis application.In addition,this paper discussed the challenges in the current community discovery field.This paper try to draw a clearer and more comprehensive outline for the community detection research field,and provide a good guide for beginners in the community detection.

Large scale Internet of Things (IoT) applications in the 5G era pose sever challenges on the network architecture in terms of heterogeneity,scalability,mobility and security.Due to the identification and location overloading problem of IP,TCP/IP based network architecture appears inefficient in addressing the challenges mentioned above.Named Data Networking (NDN) makes named content as the primary sematic and has consistency in logical topologies between network layer and application la-yer.The advantages of NDN in addressing these four challenges are reflected in the fact that naming shields the underlying hete-rogeneity,end-to-end decoupling and network layer caching provide native support for many-to-many communication and multicast,consumer mobility is supported natively by consumer driven communication pattern and content-based security is more lightweight.In this paper,future research directions of NDN based IoT were summarized.Especially,the combination of NDN and technologies including edge computing,blockchain and Software Defined Networking (SDN) to construct edge storage and computing model,centralized and distributed control model,distributed security model were proposed.

The fluid system,including the urban water supply pipe networks and the natural gas supply pipe networks,is an infrastructure with important economic and social values.They have the characteristics of wide geographical distribution,complex structure,large scale and difficult to detect.When leakage,pallution and other abnormalities occur,it's difficult to quickly find and accurately locate.With the development of sensor technology,communication technology,micro-electromechanical technology,using wireless sensor networks to monitor the system has become a research hotspot.Due to difficult communication in the fluid system,the data is difficult to transmit to the user in real time after being monitored,and can only be temporarily stored on the sensor node (referred to as a node),waiting for a suitable moment to upload.However,nodes have the characteristics of small size and easy damage,small storage capacity,weak communication capability,and limited energy.How to reliably store large amounts of data is a difficult problem.At present,some work has been done on this issue.In order to understand the progress of research in this field,this paper have carried out detailed analysis,comparison,summarization and summary of related work,and learned their advantages and disadvantages,and discussed the future research direction.

Multimedia services,especially online video services,are explosively developing. D2D(Device-to-Device) multicast content sharing is considered as a key technology that can handle massive data delivery.However,most of the current researches on D2D multicast content sharing focus on how to improve the energy efficiency of the system,while there are few researches on the data rate sum of the system,which is an important index to reflect whether the system can efficiently distribute content.In order to establish a user model which is closer to the actual scene and implement efficient content delivery to alleviate the burden of Base Stations and improve the utilization efficiency of resources (spectrum and energy),this paper proposes a kind of D2D multicast content sharing scheme based on physical-social awareness and pay incentive.Firstly,D2D multicast communication is mode-led according to the limitations of the actual scene,and the application scene of the model is expanded to the “hot spot” area with content sharing at high data rate where people are concentrated and the “blind spot” area at which the data cannot be easily transmitted directly by Base Stations in earthquake relief operations.Then,in order to effectively reduce the load of Base Stations and to cope with huge amounts of data delivery,this paper puts forward the optimization problem that the system equivalent data rate sum is regarded as an objective function under multiple constraints.In the objective function,the payment mechanism is introduced to encourage users to provide shared content for other users as cluster heads,and social ties based on similarity of interest are introduced to reduce user payment cost and improve resource utilization efficiency.Finally,a cluster head selection-cluster formation algorithm is proposed to solve this problem.In the cluster head selection algorithm,social ties based on similarity of user interest is introduced while considering the limit of user data rate threshold.In the algorithm of cluster formation,a coalition formation game of centralized control is adopted,in which the gain definition is highly consistent with the connotation of “coalition”.The simulation results show that the performance of the proposed scheme on the equivalent data rate sum and actual data rate sum is significantly improved compared with the relevant similar scheme,and it is also proved that the proposed scheme issuita-ble for large-scale user networks.

In wireless sensor network communication,sensor data need to be sent to the host computer through the wireless device.With the increase of the amount of data needed to be transmitted by the terminal sensors,the energy consumption of wireless devices is gradually increasing.A complex environment that is not convenient for timely maintenance lead to premature failure of wireless communication equipment and communication interruption.Therefore,it is necessary to compress the data collected by the sensor to reduce the amount of data sent.Based on the analysis of sensor data characteristics and traditional LZW (Lempel-Ziv-Welch) compression algorithm,an improved LZW algorithm for wireless sensor network applications is proposed.Firstly,the algorithm preprocesses the adjacent data collected from sensor to calculate the difference,so as to improve the repetition rate of the data items.Then,the appropriate dictionary size is selected and the traditional order memory is replaced with the hash memory in the dictionary,so as to improve the way of dictionary updating.When the compression rate is decreased,the proposed algorithm updates the dictionary,and saves the common single character to release the dictionary space,for data compression.The experimental results show that compared with traditional LZW algorithm,the improved LZW algorithm reduces the compression rate of the ordered sensor data by up to 40%,and reduces the amount of data needed to send.The compression speed is also increased by nearly ten times,which proves that the improved LZW algorithm for wireless sensor network applications is effective and feasible.

Direction-of-arrival (DOA) estimation based on two-dimensional planar sparse array is increasingly important in the application of massive MIMO arrays of 5G.The gridless sparse reconstruction technology promotes the development of DOA estimation research,and the super-resolution of DOA estimation methods has been advanced with the atomic norm theory.In this paper,DOA estimation is studied when spectrally-sparse signals from multiple directions are incidented on a two-dimensional sparse array.In order to accurately identify the azimuth and elevation angles of all incident signals in pairs,a two-dimensional atomic norm approach based on multiple measurement vectors (MMV) is proposed,and can be solved by semidefinite programming.The proposed algorithm extends compressive sensing of two-dimensional DOA estimation from a single measurement vector to multiple measurement vectors,so as to effectively use the joint sparsity of MMV.Numerical simulation results show that,as the MMV vector grows,the number of identifiable sources increases,the proportion of physical sensors in the sparse array decreases to 30%,the DOA estimation error decreases significantly,and the proposed algorithm can achieve a good convergence effect when the signal-to-noise ratio increases.

With the rapid development of software defined network (SDN) technology,the Internet will be in the hybrid SDN network where the traditional network devices and SDN devices coexist for a long time.It is a key scientific problem to studyenergy efficient algorithm in hybrid SDN networks.Therefore,this paper proposes a hybrid software defined network energy efficient routing algorithm based on genetic algorithm (EEHSDNGA).This paper is devoted to solving two problems.Firstly,how to choose some traditional network devices to upgrade to SDN devices in network.Secondly,how to shut down links.This paper employs genetic algorithm to solve the first problem.To solve the second problem,this paper proposes a link criticality model,which closes the links in the network one by one according to the importance of the links.The experimental results show that the energy saving ratio of EEHSDNGA in Abilene network is 36%,and in GEANT network is 42.5%.The energy saving ratio of EEHSDNGA is better than that of LF,HEATE and EEGAH.

The traditional algorithm has poor performance and its optimization solution cannot meet the diversity needs of users,when deals with the task scheduling in the cloud environment.Based on three optimization goals:task completion time,completion cost,and resource idle rate,this paper simulates the process of heuristic algorithm (the initialization,fitness assessment,task scheduling and selection stages) to construct a hierarchical evaluation and dynamic selection model(HEDSM).In the initialization phase,in order to ensure that tasks have a certain priority,the workflow task model is preprocessed using the traditional table scheduling algorithm (HEFT).In the fitness assessment phase,in order to meet the need of two aspects,the difficult solution evaluation modelsare constructed from two levels which are cloud users and cloud service providers.In the task scheduling phase,two-step scheduling is set.First,the policy set is setting,the task is pre-scheduled to ensure that the pre-scheduling scheme inherits the scheduling advantages of each strategy.Second,in order to enhance the performance of the algorithm,the task migration policy is setting to process the pre-scheduling plan.In the selection phase,the appropriate scheduling scheme is selected in the solution set according to the evaluation modle.The experiment uses WorkflowSim simulation platform and scientific workflow instance to make comparative analysis.Traditional Min-Min,Max-Min,FCFS scheduling strategies and existing IMax-Min and LWRound_Robin scheduling strategies are as comparison algorthms.The algorithms are evaluated from the diversity of user requirements and the IROS two aspects.The results show that the propsed algorithm improves the complete time and cost,therefore it is more suitable for the complex task scheduling in cloud environment.

With the advancement of communication technology,resource-constrained mobile terminal devices have been unable to meet the rapidly increasing demand for data processing by mobile users.On the one hand,mobile edge computing can be processed by migrating tasks on the mobile device to the edge computing server,which can solve the problem of insufficient computingpower of the mobile device to some extent.On the other hand,how to maintain high service performance during task migration as well as reducing the energy consumption of mobile terminals is also a topic of concern for researchers and mobile users.This paper focuses on the study of the problem of minimizing the average energy consumption of data migration based on the migration time benefit.Firstly,migration rate threshold of edge computing server detected periodically by mobile terminal is obtained by migration time revenue formula.Secondly,the optimal stopping problem of minimizing the average energy consumption of data migration with time-return constraint is constructed.It is proved that there is an optimal stopping rule and the optimal average energy consumption of data migration is obtained.Finally,the mobile terminal selects the edge computing server for task migration together with the obtained migration rate threshold and the optimal data migration average energy consumption,thereby implementing a task migration strategy with energy optimization.In the simulation experiment,the optimization strategy and other migration strategies proposed in the paper are compared on the performance parameters such as the average migration data,the average migration time,and the average data migration energy consumption.The experimental results show that compared with the other two comparison strategies,the task migration strategy with energy optimization has shorter migration time and smaller average data migration energy consumption.In addition,the performance of the effective data mobility parameter can also achieve about 10% to 40% performance improvement,and obtain better migration performance improvement effect.

With the rapid development of software defined network,deployment of the inherent defects of single controller gra-dually revealed,multiple controller deployment has become an inevitable trend.However,the number and location of controllers have a decisive influence on network performance,and the high complexity of weighing factors in solving this problem,which seriously hinders the application of SDN in data centers and wide area networks.Firstly,the essence of placement problem and its general solving steps are described.Secondly,based on the network model,the core components of the deployment strategy,namely the optimization objectives and search algorithm are described in detail.Then,based on the research at home and abroad,the deployment strategies are divided into static deployment and dynamic deployment,and the advantages and disadvantages of typical strategies are compared.Finally,the future research direction is prospected.

With the growth of the data volume for cloud applications,more and more cloud service providers pay attention to inter-datacenter bulk transfer.The main challenge of inter-datacenter bulk transfer is how to find the best resource scheduling algorithm,which uses the least resources to transfer the user’s data to the specified destinations before the specified deadline.This paper proposes MSTB(Multi-Source Tree-Based) algorithm,an effective scheduling solution for deadline-aware P2MP inter-da-tacenter transfers.With the help of multi-source mechanism and multicast forwarding tree,MSTB outperforms the state-of-the-art method.Simulation experiments show that MSTB can increase the number of transfer requests accepted by up to 91% and increase effective throughput by up to 54% with short transfer completion time and low computation complexity.

In order to reduce the communication overhead of distributed deep learning system,compression of gradient data before transmission is an effective method,such as 2Bit method in MXNet.However,there is a problem in this kind of method,that is,too high compression ratio will lead to decline in accuracy and convergence speed,especially for larger network models.To address this problem,a new gradient compression strategy called 4Bit is proposed.Four bits are used to represent a specific gradient value.Compared with 2Bit,this method can approximate the gradient more finely,thus improving the accuracy of training results and convergence speed.Furthermore,different approximation thresholds are selected according to the gradient characteristics of each layer of the network model,which makes the compressed values more reasonable,and finally improves the convergence speed and final accuracy of the model.The experimental results show that,although 4Bit is slightly lower than the 2Bit method in terms of acceleration,its accuracy is higher and practicability is better by using more bits and multiple thresholds.It is very meaningful to reduce the communication overhead of the distributed deep learning system while maintaining high accuracy by 4Bit.