In recent years, many optimization proposals for TCP BBR have been introduced, but most rely mainly on delay variations and do not fully resolve BBR’s limitations in RTT fairness, link utilization, and delay control in networks. This paper proposes BBR with Extended State (BBR-ES), which extends BBR’s state machine with a short stabilization state and a trend-based transition mechanism that react to per-flow bandwidth and RTT evolution instead of global delay alone. BBR-ES uses lightweight bandwidth and RTT trend tracking to adjust its sending rate while preserving BBR’s model-based design. Experiments on both emulated (Mininet) and real-world Internet paths (Amazon EC2) show that BBR-ES consistently improves RTT fairness and link utilization over BBRv1, BBRv3, and CUBIC while keeping queuing delay moderate and bounded; in most settings, it achieves Jain’s fairness index above 0.9 and link utilization above 98%. These results indicate that BBR-ES is a practical candidate for deployment in large-scale content delivery and a useful design reference for future model-based congestion control schemes.

In this paper, we propose a compression method of information for remote network management, such as command-response messages and operation logs of network devices. The key technologies of the proposed method are tree-type, context-aware compression dictionary, and dictionary synchronization between local and remote systems. We evaluate the performance of the proposed method using command-response history and operation logs of actual network devices. The results show that when the dictionary used for compression is constructed offline, 1-7% of the compression ratios1are obtained regardless of the used dataset. When the compression dictionary is updated online, the compression ratio degrades but remains far superior to conventional methods. We also confirm that pre-processing target information can significantly improve the compression ratio.

A main factor that hindered the performance of most congestion control algorithms in the current Internet is the lack of observable information about the network congestion situation. The most cost-effective and least complex way of tackling the problem of congestion detection is to implement a local solution based on the information gathered on the sender. The number of competing flows on the network bottleneck is one of the indicator of the network congestion situation. In this paper, we explore the possibility of estimating the number of competing flows in a network bottleneck using machine learning trained on locally observable information. We used the time series classification-based method and regression-based method to perform the estimation. The proposed method is evaluated in a virtualized network. The trained classification model results in an accuracy of 1.0 when the number of competing flows it encounters is trained in advance. And the regression method can have a MAPE of around 30 [%] in most situations for the number of competing flows that have not been trained into the model.

Vehicular mobility and connectivity vary significantly over space and time when vehicular crowd sensing covers a city-wide area for a long time period, but it is important to achieve sufficiently uniform data coverage to satisfy the requirements of an environmental monitoring scenario. Our goal is thus to ensure uniform spatial-temporal coverage of sensed data over a city-wide area despite such vehicle dynamics. For a large area, trajectory-based approaches must deal with a great number and variety of participant mobility patterns. Hence, we propose a probabilistic control mechanism that adaptively adjusts the incentive to each participant, without using any prior information about participants. We provide a mathematical analysis that ensures stability of the number of participants with assigned tasks (called workers), and we evaluate the mechanism's robustness by using 24-hr vehicle trace data from a city-wide area. Our results demonstrate that, when the number of participants is up to 1500 times higher than the required number of workers, sensing actions result in a distribution with a mean of about 1 and an interquartile range of around 4 for a required sensing interval; moreover, the mean increases by 2% when 30% of communication messages are randomly lost.

In recent years, there’s been a surge in the popularity and affordability of distributed power generation equipment, such as photovoltaic systems (PV) and energy storage systems. At present, however, most solutions target individual user’s energy management. Given the varied energy consumption habits, networking neighboring users and managing energy as a unified system could boost efficiency. Yet, this comes with challenges: unpredictable dynamic demands, significant computational loads, and concerns over data privacy. To tackle these challenges, we introduce a management system that merges deep reinforcement learning (DRL) with federated learning (FL) techniques, named PDDPG-FL. In this setup, each home possesses an agent responsible for decisions like charging/discharging and trading energy with other users. For every agent, we employ a priority-aware deep deterministic policy gradient (PDDPG) algorithm. This not only addresses fluctuating demand adeptly but also offers computational advantages over the conventional DDPG algorithm. Moreover, by incorporating the FL framework, agents can collaborate without risking data privacy breaches. Simulation results show that PDDPG-FL can reduce dependency on main supply grids by up to 9.7% and offers a more streamlined computational process.

For the future cyber-physical system (CPS) society, it is necessary to construct digital twins (DTs) of a real world in real time using a lot of cameras and sensors. Hence, the energy efficiency of both networks and computers for largescale distributed video analysis is a major challenge for the full-scale spread of CPSs and DTs. Toward this goal, we first propose a model to arbitrarily split and distribute the video analysis task to terminals, edge servers, and cloud servers and dynamically assign appropriate CNN models to them. System-wide optimization of such distributed processing can reduce overall system power consumption by reducing network bandwidth and efficiently utilizing distributed CPU/GPU resources. To realize this optimization in a real system, we also propose a model to estimate the GPU load, processing time, and power consumption of these devices based on massive experimental measurements. Since such a large-scale optimization is difficult because of the dynamic and multi-objective nature of the problem, we propose a new optimization algorithm composed of Genetic Algorithm and Bayesian Attractor Model. Finally, simulation evaluations are performed to demonstrate that the proposed method can minimize system power consumption and satisfy latency and recognition accuracy requirements of each video analysis, even under changing environmental conditions.

In recent years, the fast development of mobile communications and cloud systems has substantially promoted edge computing. By pushing server resources to the edge, mobile service providers can deliver their content and services with enhanced performance, and mobile-network carriers can alleviate congestion in the core networks. Although edge computing has been attracting much interest, most current research is application-specific, and analysis is lacking from a business perspective of edge cloud providers (ECPs) that provide general-purpose edge cloud services to mobile service providers and users. In this article, we present a vision of general-purpose edge computing realized by multiple interconnected edge clouds, analyzing the business model from the viewpoint of ECPs and identifying the main issues to address to maximize benefits for ECPs. Specifically, we formalize the long-term revenue of ECPs as a function of server-resource allocation and public data-placement decisions subject to the amount of physical resources and inter-cloud data-transportation cost constraints. To optimize the long-term objective, we propose an online framework that integrates the drift-plus-penalty and primal-dual methods. With theoretical analysis and simulations, we show that the proposed method approximates the optimal solution in a challenging environment without having future knowledge of the system.

In future beyond 5G networks, the demand on the network would be more diversified and personalized. The variety of underlying physical network would also be accelerated by utilization small and heterogeneous cells and the co-existence of cellular networks and other types of wireless networks. User mobility and unexpected application service utilization would highly fluctuate the network demand spatio-temporally. However, per-service and static network slicing considered in 5G networks cannot accommodate such requirements. In this paper, we propose a novel network architecture, User-Oriented Network slicing Architecture (UONA). The proposed architecture has the following two major characteristics. One is that we maintain network slices on a per-user basis, not on a per-service basis. Such high-resolution of network slices would satisfy various and personal requirements from users. The other is decoupling the process of generating network slices into two subprocess, providing subslices by subslice providers and constructing end-to-end user network slices from the subslices by network slice brokers.We introduce the overall design of UONA and explain its advantages, as well as the research challenges to realize it. We also demonstrate the effectiveness of per-user configuration of network subslice by presenting the preliminary numerical evaluation results.

One of the major shortcomings of Centralized Radio Access Networks (C-RAN) is that the large capacity is required for fronthaul network between Remote Radio Heads (RRHs) and central office with baseband unit (BBU) pool. Possible solutions are to introduce lower-cost networking technology for fronthaul network, such as Time and Wavelength Division Multiplexing Passive Optical Network (TWDM-PON), and to introduce functional split, that moves some baseband processing functions to cell site to decrease the utilization of the fronthaul network. In this paper, we construct the mathematical model for selecting function split options of baseband processing to minimize the power consumption of TWDM-PON based fronthaul network. In detail, we formulate the optimization problem for minimizing the total power consumption of fronthaul network in terms of the capacity of TWDM-PON, the number of RRHs in each cell site, server resources, latency constraints, the amount of traffic from each RRH, physical/virtual server power consumption characteristics. Numerical examples are shown for confirming the correctness of the proposed model and for presenting the effect of resource enhancement methods on the capacity and energy efficiency of the system.

In recent years, the rapid development of cloud computing, networking, and mobile computing have substantially promoted mobile edge computing (MEC). Currently, most of the MEC services can be roughly divided into two categories: computation offloading to accelerate computation and save the energy of mobile devices and data services to shorten the latency between the content providers and the mobile users. Although emerging services such as user-specified transcoding and AR/VR systems require joint optimization of computing resource allocation and data placement, there is little research on it. In this work, we carry out an in-depth study on the interaction of computing resource allocation and data placement in mobile edge computing environments. Based on the analysis of the temporal and spatial characteristics of the two tasks, we propose a joint optimization framework that works with online manner. The proposed method employs hybrid timescales: a coarse-grained timescale to update the data placement and a fine-grained timescale to decide computing resource allocation. The proposed method achieves provable near-optimal performance without buffering users' requirements and does not assume that future trends in user requirements are predictable.

In response to the growing demand for cellular networks, it is essential to improve the capacity of mobile core networks. Especially, in terms of accommodating machine-to-machine/Internet-of-Things (M2M/IoT) terminals into cellular networks, the load on the control and the user planes of the mobile core network increases massively. To deal with this problem, it is possible to apply virtualization technologies, such as software-defined network and network function virtualization. However, few existing studies evaluate such solutions for mobile core networks numerically and in detail. In this paper, we first evaluate mobile core network architectures with virtualization technologies and control/user (C/U) plane separation using the mathematical analysis. We also propose a novel bearer aggregation method to reduce the control plane load to accommodate massive M2M/IoT terminals. The result of numerical evaluation shows that the capacity of the mobile core network can be increased by up to 32.8% with node virtualization and C/U plane separation, and further by 201.4% by using bearer aggregation. Moreover, to maintain the performance of the mobile core network, we should carefully determine where the bearer aggregation is applied and when the shared bearer for each terminal is determined based on application characteristics and the number of accommodated M2M/IoT terminals.

In this paper, we propose a novel energy-efficient architecture for software-defined data center infrastructures. In our proposed data center architecture, we include an exhaust heat reuse system that utilizes high-temperature exhaust heat from servers in conditioning humidity and air temperature of office space near the data center. To obtain high-temperature exhaust heat, equipment such as server racks and air conditioners are deployed in tandem so that the aisles are divided into three types: cold, hot, and super-hot. In this paper, to investigate the fundamental characteristics of our proposed data center architecture, we consider various types of data center models and conduct numerical simulations that use results obtained by experiments at an actual data center. Through simulation, we show that the total power consumption by a data center with our proposed architecture is 27 percent lower than that by data center with a conventional architecture. In addition, it is also shown that the proposed tandem equipment arrangement is suitable for obtaining high-temperature exhaust heat and decreasing the total power consumption significantly under a wider range of conditions than in the conventional equipment arrangement.

Modern webpages consist of many rich objects dynamically produced by servers and client terminals at diverse locations, so we face an increase in web response time. To reduce the time, edge computing, in which dynamic objects are generated and delivered from edge nodes, is effective. For ISPs and CDN providers, it is desirable to estimate the effect of reducing the web response time when introducing edge computing. Therefore, in this paper, we derive a simple formula that estimates the lower bound of the reduction of the response time by modeling flows obtaining objects of webpages. We investigate the effect of edge computing in each webpage category, e.g., News and Sports, using data measured by browsing about 1,000 popular webpages from 12 locations in the world on PlanetLab.

To reduce the power consumption in data centers, the coordinated control of the air conditioner and the serversis required. It takes tens of minutes for changes of operationalparameters of air conditioners including outlet air temperatureand volume to be reflected in the temperature distribution inthe whole data center. So, the proactive control of the airconditioners is required according to the prediction temperaturedistribution corresponding to the load on the servers. In thispaper, the temperature distribution and the power efficiencyof air conditioner were predicted by using a machine-learningtechnique, and also we propose a method to follow-up proactivecontrol of the air conditioner under the predicted optimumcondition. Consequently, by the follow-up proactive control ofthe air conditioner and the load of servers, power consumptionreduction of 30% at maximum was demonstrated.

Network tomography is a promising technique to achieve fault management in networks where the existing IP-based troubleshooting mechanism cannot be used. Aiming to apply Boolean network tomography to fault management, various heuristic methods for configuring monitoring trails have been proposed to localize link failures in all-optical mesh networks. However, these existing heuristic methods must be executed in a centralized server that administers the entire managed network, and present scalability problems when they are applied to large-scale managed networks. Thus, this paper proposes a novel scheme for achieving lightweight Boolean network tomography in a decentralized manner. The proposed scheme partitions the managed network into multiple management areas and localizes link failures independently within each area. This paper also proposes a heuristic network partition method with the aim of implementing the proposed scheme efficiently. The effectiveness of the proposed scheme is verified using a typical fault management scenario, where all the single-link failures are localized by the monitoring paths the routes of which are predetermined. Simulation results show that the proposed scheme can greatly reduce the computational load on the fault management server when Boolean network tomography is deployed in large-scale managed networks.

Web performance, the time from clicking a link on a web page to finishing displaying the web page of the link, is becoming increasingly important. Low web performance of web pages tends to result in the loss of customers. In our research, we measured the time for downloading files on popular web pages by running web browsers on four hosts worldwide using PlanetLab and detected the longest portion in download time. We found the longest portion in download time to be Blocked time, which is the waiting time for the start of downloading in web browsers. In this paper, we propose a method for accelerating web performance by reducing such Blocked time with a cache of rendering results. The proposed method uses an in-network rendering function which renders web pages instead of web browsers. The in-network rendering function also stores the rendering results in its cache and reuses them for other web browsers to reduce the Blocked time. To evaluate the proposed method, we calculated the web performance of web pages whose render results are cached by analyzing the measured download time of actual web pages. We found that the proposed method accelerates web performance of long round trip time (RTT) web pages or long RTT clients if the web pages' dynamic file percentages are within 80%.

To decrease the power consumption of data centers, coordinated control of air conditioners and task assignment on servers is crucial. It takes tens of minutes for changes of operational parameters of air conditioners including outlet air temperature and volume to be actually reflected in the temperature distribution in the whole data center. Proactive control of the air conditioners is therefore required according to the predicted temperature distribution, which is highly dependent on the task assignment on the servers. In this paper, we apply a machine learning technique for predicting the temperature distribution in a data center. The temperature predictor employs regression models for describing the temperature distribution as it is predicted to be several minutes in the future, with the model parameters trained using operational data monitored at the target data center. We evaluated the performance of the temperature predictor for an experimental data center, in terms of the accuracy of the regression models and the calculation times for training and prediction. The temperature distribution was predicted with an accuracy of 0.095°C. The calculation times for training and prediction were around 1,000 seconds and 10 seconds, respectively. Furthermore, the power consumption of air conditioners was decreased by roughly 30% through proactive control based on the predicting temperature distribution.

In this paper, we propose a method for increasing the capacity of Machine-To-Machine (M2M) communication in mobile core networks. The proposed method combines two approaches: bearer aggregation inside mobile core networks for decreasing the load of Evolved Packet Core (EPC) nodes, and applying a Software Defined Networking (SDN) architecture to separate the control and data planes and aggregate control plane nodes in a cloud network environment for resource sharing. The combination of these two approaches is meaningful because they have a complementary relationship. We give a mathematical analysis and numerical results of a performance evaluation of the proposed method. The evaluation results show that we can increase the capacity of a mobile core network for M2M communication by around 30% when one of the two approaches is applied, while the performance gain increases up to 124% when both approaches are combined.

Modern websites consist of many rich objects dynamically produced by servers and client terminals at diverse locations. Consequently, we face complications in understanding the communication structure generated when accessing websites. To reduce the response time at browsed websites, many website objects are delivered using content delivery networks (CDNs), in which data objects are delivered from cache servers located close to user terminals. Although the use of CDNs have been assumed to reduce web response time, the actual effect of CDNs on this reduction has not been clarified. To answer this fundamental question, we measured the communication structure of traffic generated when accessing the 1,000 most popular websites from 12 locations worldwide. We found, for example, that it will be desirable to give high priority to entertainment websites at night and to business-related websites during the day.

Machine Type Communications (MTC) has been paid much attention as a new communication paradigm to increase mobile network traffic. Most of MTC terminals are time-driven, that is, they send and receive data periodically. Therefore, network access requests on mobile core networks are concentrated at a specific timing, which results in instantaneous increase in network load. Considering the fact that such time-driven MTC would accept a certain amount of latency in their cyclic communication, in this paper, we propose a scheduling method of communication timings of time-driven MTC terminals to mitigate traffic concentration. We extend the standardized back-off mechanism of 3GPP to configure the back-off time length for each terminal to decrease the number of concurrent bearers in the network, while satisfying requirements on communication latency. We compare proposed methods by simulation experiments and reveal that we can achieve almost zero access rejections at reasonable communication quality by a simple timeslot selection algorithm when the core network maintain the timeslot assignment status for accommodated User Equipments. To the best of our knowledge, this is the first proposal to alleviate short-term congestion of mobile core networks by MTC with TDMA-like network control.

Existing techniques for measuring available bandwidth measure the available bandwidth at bottlenecks along the path, and most of them do not specify the bottleneck location. In this paper, we propose an end-to-end measurement method for the hop-by-hop available bandwidth along a network path. Such a technique can facilitate advanced traffic control, especially in heterogeneous network environments. The proposed method assumes a situation where intermediate routers can record the arrival and departure times of incoming packets as timestamps in the packets themselves. The end host sends probe packets at various rates and estimates the available bandwidth at each network section using the incoming and outgoing rates of packets calculated from intermediate timestamps, based on statistical processing under a fluid traffic model. We present extensive simulation results for the proposed method and confirm that it can accurately measure the available bandwidth of each section along the network path even when the available bandwidth of the sender-side network is smaller than that of the receiver-side network.

In a multitenant data center, nodes and links of tenants' virtual networks (VNs) share a single component of the physical substrate network (SN). A failure of the single SN component can thereby cause simultaneous failures of multiple nodes and links in a VN, this complex of failures must significantly disrupt the services offered on the VN. In the present paper, we clarify how the fault tolerance of a VN is affected by a SN failure, especially from the perspective of VN allocation in the SN. We propose a VN allocation model for multitenant data centers and formulate a problem that deals with the bandwidth loss in the VN due the SN failure. We conduct numerical simulations with the setting that has 1.7 × 108bit/s bandwidth demand on each VN. The results show that the bandwidth loss can be reduced to 5.3 × 102bit/s per VN, but the required bandwidth between physical servers in the SN increases to 1.0 × 109bit/s per VN when each node in the VN is mapped to an individual physical server. The balance between the bandwidth loss and the required bandwidth between physical servers can be optimized by assigning every four nodes of the VN to each physical server, meaning that we minimize the bandwidth loss without providing too sufficient bandwidth in the core area of the SN.

Energy efficiency of wireless clients is an important issue in wireless communications. When multiple network applications are running concurrently on a single wireless client, packets belonging to each application are sent and received independently, but these packets are multiplexed at the MAC level. This uncoordinated behavior makes it difficult to control sleep timing. In addition, transitioning between active and sleep modes frequently will consume a non-negligible amount of energy. To address these difficulties, we have proposed SCTP tunneling: a transport-layer approach that resolves the coordination problem and so reduces the energy consumed by multiple TCP flows on a wireless LAN (WLAN) client. In this study, we perform experiments with off-the-shelf WLAN devices to assess the energy efficiency and the transfer time of SCTP tunneling. We show experimentally that SCTP tunneling with unscheduled automatic power save delivery (U-APSD) can save energy when compared to U-APSD alone, while still increasing file transfer speed.

Multihoming is widely used by Internet service providers (ISPs) to obtain improved performance and reliability when connecting to the Internet. Recently, the use of overlay routing for network application traffic is rapidly increasing. As a source of both routing oscillation and cost increases, overlay routing is known to bring challenges to ISPs. In this paper, we study the interaction between overlay routing and a multihomed ISP's routing strategy with a Nash game model, and propose a routing strategy for the multihomed ISP to alleviate the negative impact of overlay traffic.We prove that with the proposed routing strategy, the network routing game can always converge to a stable state, and the ISP can reduce costs to a relatively low level. From numerical simulations, we show the efficiency and convergence resulting from the proposed routing strategy. We also discuss the conditions under which the multihomed ISP can realize minimum cost by the proposed strategy.

We propose a TCP window-size delegation method for TXOP Exchange applicable to the downlink in wireless access networks. In TXOP Exchange, the compliant stations (STAs) cooperatively use their available bandwidth in accordance with their required QoSs. TXOP Exchange was previously validated for the uplink. The proposed delegation method enables STAs to delegate their bandwidth for the downlink as well without requiring any modifications to the legacy access point or the STAs. Simulation demonstrated that this method works well.

This paper proposes a novel recovery mechanism from large-scale network failures caused by earthquakes, terrorist attacks, large-scale power outages and software bugs. Our method, which takes advantage of overlay networking technologies, pre-calculates multiple routing configurations to prevent possible simultaneous network failures and selects one configuration immediately after detecting the failures. Through numerical calculation results using actual AS-level topology, we show that our proactive method improves network reachability from 89% to 99%, while keeping the path length sufficiently short, when up to 8% of the nodes in a network are down simultaneously.

In this paper, we propose a new mechanism which detects tampered-TCP connections at edge routers and protects well-behaved TCP connections from the tampered-TCP connections, resulting in maintaining the fairness amongst TCP connections. The proposed mechanism monitors the TCP packets at an edge router and estimates the window size or the throughput for each TCP connection. By using estimation results, the proposed mechanism assesses whether each TCP connection is tampered or not and drops packets intentionally if necessary to improve the fairness amongst TCP connections. From the results of simulation experiments, we exhibit that the proposed mechanism can accurately identify tampered-TCP connections. We also show that the proposed mechanism can regulate throughput ratio between tampered-TCP connections and competing TCP Reno connections to about 1.

TCP overlay networks that control data transmission quality at the transport layer are being paid a lot of attention as users' demands for diversified Internet services increase. They are expected to enhance the end-to-end throughput of the TCP connection essentially because the round trip times and the packet loss ratios of each split TCP connection are reduced. However, performance degradation may occur due to undesired interactions among the split TCP connections. We introduce an analysis approach to estimate end-to-end throughput of data transmission with a TCP proxy mechanism considering performance degradation. Our analysis results reveal that we confirm the effect of the TCP proxy mechanism. We also find that we cannot ignore performance degradations due to interactions among split TCP connections, especially when the congestion level of the network they traverse is small. Further, we clarify that we should take into account the packet loss ratios, performance degradations and propagation delays of the network when we consider issues relating to the design of TCP overlay networks.

Appropriate control parameters are important for the successful deployment of RED (Random Early Detection) routers, especially when many TCP connections share the bottleneck link. In this paper, we first describe a new simple analysis method for determining the window size distribution of many TCP connections sharing a bottleneck router. We consider two kinds of buffering disciplines: TD (Tail Drop) and RED. We model the window size evolution of TCP connections by using a Markov process whose state is represented by a set of the current window size and the ssthreth value. The state transition matrix is then calculated by considering the characteristics of TD and RED routers. We show numerical results demonstrating the accuracy of our analysis and we discuss the fairness of TD and RED. We confirm that RED does not help improve the router's throughput even when appropriate control parameters are chosen but that it is still useful to provide the fairness among many competing TCP connections.

Although many research efforts have been devoted to network congestion in the face of an increase in the Internet traffic, there is little discussion on performance improvements for endhosts. We propose a new architecture, called scalable socket buffer tuning (SSBT), to provide high-performance and fair service for many TCP connections at Internet endhosts. SSBT has two major features. One is to reduce the number of memory accesses at the sender host by using some new system calls, called simple memory-copy reduction (SMR) scheme. The other is equation-based automatic TCP buffer tuning (E-ATBT), where the sender host estimates 'expected' throughput of the TCP connections through a simple mathematical equation, and assigns a send socket buffer to them according to the estimated throughput. If the socket buffer is short, the max-min fairness policy is used. We confirm the effectiveness of our proposed algorithm through both a simulation technique and an experimental system. From the experimental results, we have found that our SSBT can achieve up to a 30% gain for Web server throughput, and a fair and effective usage of the sender socket buffer can be achieved.

According to past research, a TCP Vegas version is able to achieve higher throughput than TCP Tahoe and Reno versions, which are widely used in the current Internet. However we need to consider a migration path for TCP Vegas to be deployed in the Internet. In this paper, by focusing on the situation where TCP Reno and Vegas connections share the bottleneck link, we investigate the fairness between two versions. From the analysis and the simulation results, we find that the performance of TCP Vegas is much smaller than that of TCP Reno as opposed to an expectation on TCP Vegas. The RED algorithm improves the fairness to some degree, but there may still be an inevitable trade-off between fairness and throughput. Accordingly, we consider two approaches to improve the fairness. The first one is to modify the congestion control algorithm of TCP Vegas, and the other is to modify the RED algorithm to detect misbehaved connections and drop more packets from those connections. We use both of analysis and simulation experiment for evaluating the fairness, and validate the effectiveness of the proposed mechanisms.

We investigate the performance of TCP under three representatives of packet scheduling algorithms at the router. Our main focus is to investigate how fair service can be provided for elastic applications sharing the link. Packet scheduling algorithms that we consider are FIFO (first in first out), RED (random early detection), and DRR (deficit round robin). Through simulation and analysis results, we discuss the degree of achieved fairness in those scheduling algorithms. Furthermore, we propose a new algorithm which combines RED and DRR algorithms in order to prevent the unfairness property of the original DRR algorithm, which appears in some circumstances where we want to resolve the scalability problem of the DRR algorithm. In addition to the TCP Reno version, we consider TCP Vegas to investigate its capability of providing fairness. The results show that the principle of TCP Vegas conforms to DRR, but it cannot help improving the fairness among connections in FIFO and RED cases, which seems to be a substantial obstacle for the deployment of TCP Vegas.

We focus on fairness and stability of the congestion control mechanisms adopted in several versions of TCP by investigating their time-transient behavior through an analytic approach. In addition to TCP Tahoe, TCP Reno, and TCP Vegas, we consider enhanced TCP Vegas which is proposed in this paper for fairness enhancements. We consider the homogeneous case, where two connections have the equivalent propagation delays, and the heterogeneous case, where each connection has different propagation delay. We show that TCP Tahoe and TCP Reno can achieve fairness among connections in the homogeneous case, but cannot in heterogeneous case. We also show that TCP Vegas can provide almost fair service among connection, but there is some unfairness caused by the essential nature of TCP Vegas. Finally, we explain the effectiveness of our enhanced TCP Vegas in terms of fairness and throughput.