Real-time and reliable wireless communication is increasingly vital for applications like remote robot control. To support such needs, radio environment maps (REMs) showing metrics like signal strength at each location in the area have gained attention. However, their values are affected by noise, fading, and positioning errors, so representing only the mean is insufficient; their full probability distribution is needed. While Gaussian Process Regression (GPR) is commonly used to build REMs, it often struggles to capture distribution shapes and is inefficient at scale. To address these challenges, we build on Probabilistic Feature Networks using Prior-Data Fitted Networks (PFNs), which utilize a transformer trained on GPR, and propose the Mixture of PFNs (MoPFNs) to model the probability distribution as a mixture of Gaussian distributions. Experimental results in a 28 GHz millimeter wave (mmWave) environment show that the proposed method reduces the average error in estimated mean RSRP to less than 10 dB, accurately captures variations in distribution shape across locations, and improves execution time by approximately 92.5%.

In recent years, many surveillance cameras have been installed in cities, and human tracking technology has received much attention. In most current human-tracking technologies, servers collect images of people and then analyze their features from the data. In this method, the network loads on the servers increase as the number of people tracked increases, causing problems, such as packet loss and loss of real-time performance. In this paper, we propose two real-time human tracking methods. The methods conduct a human tracking process without servers by sharing extracted human features among devices. Experimental evaluations of the amount of communication traffic and processing time using multiple cameras have shown that the two proposed methods can distribute the network load with slight deterioration in processing speed and tracking accuracy.

In Cyber-Physical Systems (CPS), the reliability of wireless communication is paramount to ensuring safety. Received Signal Strength (RSS) map is particularly beneficial for safe robot operation, for example. However, accurately estimating an RSS map poses significant challenges, particularly when higher frequency bands are employed for broadband communications. Therefore, it is also important to not only further enhance the accuracy of the estimation, which is the target of many existing methods, but also design system that can tolerate errors in the estimation. In this paper, we propose a method for constructing a digital twin that represents the quality of the wireless network using probability distributions. Representing data probabilistically proves effective for risk-sensitive robot control or robust planning of base station positioning. We propose using a graphical model known as Markov Random Field (MRF), to depict the spatial structure of the RSS map. The probability distribution of the RSS value at each point within the space is provided as the marginal probabilities of the MRF. We then evaluated the proposed method in our own 28 GHz Private 5G environment and extensively studied the characteristics of indoor millimeter radio communication. We confirmed that each estimated point can be well estimated by probability distribution using the proposed method. In addition, the design of error toler-ance based probability distribution is discussed. By determining a margin on the estimated expected value based on the standard deviation of the estimated distribution, the margin can be set more efficiently than by determining a uniform margin for each points.

To create a future society based on cyber-physical systems, we need real-time digital twins made with cameras and sensors. The challenge is making this energy-efficient. A model in [1] suggests dividing video analysis tasks among terminals, edge servers, and cloud servers to minimize power consumption. This paper addresses energy optimization in dynamic environments with a two-level approach: detecting and categorizing environmental changes (LEC and SEC) and using a two-level adaptive evolutionary algorithm (TAEA) to make corresponding adjustments. A case study with a differential evolution algorithm demonstrates its effectiveness in minimizing energy costs and improving processing accuracy and latency violations.

In recent years, the use of computer systems for identifying people has become increasingly popular. Clear facial images and detailed facial features are often provided to the system to improve identification accuracy, but if they are misused, privacy can be compromised. The authors are currently building a next-generation video blog system that can do the following: capture images of passersby with cameras, blur the facial images of all the people, remove the blurring for specific people who have registered their facial images for public viewing, and then upload them as video blogs in real time in a hands-free manner. By comparing the similarities between the obscured data in the user's stored data and the data captured by the camera and immediately obscured and sent to the edge server, the authors devised a system that enables person identification but does not pass clear facial images and features to the system. The authors focused on the possibility of using noise strength added to features and random seeding of features as a common quasi-encryption key to protect privacy. This paper assesses the extent to which the intensity of noise in this system obscures person identification. The results show that privacy can be protected by adding epsilon noise below a certain strength to the features, and the noise strength and the random number seed for noise generation can be used as a common quasi-cryptographic key.

Network slicing technology is required to dynamically provide virtual networks in response to user requirements with a wide variety of services operating on the network. Generally, optimal allocation of virtual networks to resources on the real network is a combinatorial optimization problem, and it is difficult to find an exact solution in realistic time in the current large-scale and complex networks. In addition, user requirements change dynamically, and therefore, optimization methods that can cope with such temporal variations in the situation are required. In this paper, we propose a method to solve a virtual network embedding problem using quality-diversity (QD) algorithms, especially the MAP-Elites algorithm, and evaluate its effectiveness through computer simulations.

To enable robust wireless communications towards the future Cyber Physical System by optimizing beamforming in real time, it is essential to quickly estimate changes in the radio environment map (REM) after altering the beam direction. In addition, because of the difficulty to predict detailed REMs with high accuracy, especially for millimeter waves, a probabilistic REM, which estimates the radio signal strength at any point as a probability distribution rather than as a deterministic value, is expected. In this paper, to enable fast and probabilistic REM estimation, we propose a scheme that divides the estimation formula into a linear term and a Gaussian Process Regression (GPR) term to minimize the recalculation time associated with beam adjustments. We also propose a GPR kernel that uses polar coordinates centered on the base station angle to better capture the characteristics of narrow beams like millimeter waves. We evaluated the proposed scheme using Kullback-Leibler Divergence (KLD) to compare the measured and estimated distributions of the radio signal strength. In the REM estimation after beam adjustment, an accurate distribution is estimated at 29 out of 35 points with a KLD of 0.5 or lower. The proposed method achieved approximately 50 times faster REM reconstruction compared to a simple REM using GPR, enabling dynamic beam-forming optimization in mm-wave environment.

We discuss "GPU Time Sharing," which involves sharing GPU servers locally owned in the private cloud to increase their usage and to reduce the costs of cloud services. In the GPU Time Sharing, users can reserve and use a GPU server for a certain period of time to complete a task, such as training the AI model or inference on a video. In the reservation, task completion time must be predicted conservatively to ensure that the task is completed in the reserved time, but at the same time, the reserved time should be as short as possible for efficient resource sharing. In this paper, we propose a task completion time prediction method called "Uncertainty Scaled Gradient Boosting Decision Tree" (USGBDT), which first predicts the completion time of the Deep Learning (DL) tasks using the Gradient Boosting Decision Tree, and then scales the predicted time based on the expected uncertainty of the machine learning models. Applying the proposed method to the GPU Time Sharing for video analysis tasks, we have confirmed that all tasks are completed in the predicted completion time and the GPU usage time over the reserved time is improved from 53.4% to 67.0%.

The development of Internet of Things and smart building technologies will accelerate many real-world entities (e.g., equipment, people flow, comfort levels) to be managed as digital data. This enables a Software-Defined Building in which advanced and complex applications realize a broad range of value propositions for smart buildings. To enable a scalable platform that can easily add and integrate a wide range of data and applications, the main issue addressed is the difficulty in unifying existing building equipment data models used in BACnet or others, or any related information in the building. Based on the Web of Things architecture, we propose a data model that accommodates both the information within buildings and the broad range of information expected to increase in the future. To enable the proposed data model to be efficiently managed by a building OS, we also propose a hierarchical mechanism that incorporates modules to convert the group of devices connected by different protocols into a unified data model. We develop a prototype building OS platform using Eclipse Ditto to evaluate the proposed model and mechanism. We confirm that the platform can correctly manage device information and that its performance is sufficient.

To achieve efficient human-robot cooperation, it is necessary to work in close proximity while ensuring safety. However, in conventional robot control, maintaining a certain distance between humans and robots is required for safety, owing to control uncertainties and unexpected human actions, which can limit the efficiency of robot operations. Therefore, this study aims to establish a human-robot cooperation aiding system that concerns both safety and efficiency in a close proximity situation. We propose two Augmented Reality (AR) interfaces to display robot information via AR glasses, allowing workers to see the robot information while focusing on their task and avoiding collisions with the robot. AR glasses can give hands-free communication required for a work environment like warehouses or convenience store backyards, and multiple information levels, simple or informative, to balance accuracy and easiness of human recognition ability. We conducted a comparative evaluation experiments with 24 participants and found that both safety and efficiency were improved using the proposed user interfaces (UIs). We also collected the position, head motion, and eye-tracking data from the AR glasses to gain insight into human behavior during the tasks for each UI. Consequently, we clarified the behavior of the participants under each condition and how they contributed to safety and efficiency.

In 5G, flexible resource management, mainly by base stations, will enable support for a variety of use cases. However, in a situation where a large number of devices exist, such as in mMTC, devices need to allocate resources appropriately in an autonomous decentralized manner. In this paper, autonomous decentralized timeslot allocation is achieved by using a decision model for each device. As a decision model, we propose an extension of the Bayesian Attractor Model (BAM) using Bayesian estimation. The proposed model incorporates a feature of human decision-making called magnitude sensitivity, where the time to decision varies with the sum of the values of all alternatives. This allows the natural introduction of the behavior of making a decision quickly when a time slot is available and waiting otherwise. Simulation-based evaluations show that the proposed method can avoid time slot conflicts during congestion more effectively than conventional Q-learning based time slot selection.

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.

To realize real-time mobile augmented reality applications, various objects in the real world need to be instantly identified, located, and represented as a digital twin through sensor devices and edge IoT systems. However, it is challenging to make a fast and accurate decision on what the object is from real-time noisy streaming information. Multimodal decision making has been expected to mitigate such incomplete information and improve the accuracy of simplified recognition algorithms tuned for edge devices. In this paper, we propose an object estimation method inspired from the multimodal information processing mechanism of the brain, which makes decisions based on multiple types of uncertain observed information. Through computer simulations, we show that our proposed method identifies an object accurately and quickly from uncertain observed information.

LoRaWAN has become a major research target in low-power and wide-area (LPWA) communication because of its ease of development and the possibility of building self-managed networks. In LoRaWAN, the data rate of a node can be dynamically controlled by the gateway through changing the spreading factor of nodes, and this control can be performed according to the network conditions. However, it is difficult to immediately grasp the individual states of a large number of nodes, calculate the optimal data rate, and assign the appropriate spreading factor to the nodes due to the low communication speed of LoRaWAN compared to conventional wireless networks. In this paper, we propose a spreading factor allocation method for LoRaWAN that simultaneously improves the data arrival rate and network lifetime by using the thermodynamical genetic algorithm. Through computer simulation, we show that our proposed method assigns an appropriate spreading factor to each node, which achieves a higher data arrival rate and prolongs the network lifetime.

Recently, digital twins have been paid much attention as a major application towards Beyond 5G/6G network, and real-time object recognition methods are key technology to digitize the real world as a digital twin. However, it is challenging to make a fast and accurate decision on what the object is from real-time streaming information such as video because accurate object recognition algorithms require a huge computation. To satisfy delay requirement of digital twin applications, such computations have to be moved from cloud to edges or even small terminal devices, where computing capacity is very limited. Thus, recognition mechanisms have to be simplified for small devices but they would result in degraded accuracy. In this paper, we focus on the multimodal information processing mechanism of the brain, which makes decisions based on multiple types of uncertain observed information, to improve accuracy of simplified recognition mechanisms. We first propose a unimodal object recognition mechanism based on the Bayesian attractor model, which continuously recognizes objects from noisy streaming media data. Then, we extend the mechanism with Bayesian causal inference to fuse the results of unimodal media recognition. Through computer simulations, we show that our proposed method identifies an object accurately and quickly from uncertain observed information.

The “Internet of things” has become a common term, and low-power wide-area (LPWA) technology is attracting much attention as one of its elemental technologies. LPWA achieves wide-area communication without consuming much energy, allowing various data sensing and gathering applications. LoRa is an LPWA communication technology that uses unlicensed bands. Because it is possible to build a self-managed network with LoRa, many LoRa-based services will be scattered in the same area without an overall administrator. As a result, the communication performance of LoRa may degrade due to unintended radio interference. Unfortunately, many LPWA techniques, including LoRa, have low data rates, making it difficult to gather sufficient control information to avoid such degradation of communication performance. In this paper, we propose a method for estimating network congestion states through successive estimation using Bayesian updates of prior distributions. Computer simulations show the network state can be estimated by our proposed method with accumulating a little control information.

Automatizing the process of designing communication services in network function virtualization (NFV) is important because it may reduce provisioning time and lead to more efficient designs. The design process involves solving performance constraints imposed by service level agreements (SLAs), which in turn requires accurate and fast performance prediction. However, effects such as resource contention make performance prediction in virtualized environments challenging when large numbers of possible combinations of software and hardware are considered. The key to scalability lies in finding a componentized approach that reduces the number of model degrees of freedom while still allowing high accuracy. In this work, we propose a componentized approach based on feed-forward networks that are composited from software and hardware models. Model parameter data is obtained from a machine learning technique which is fed using data generated from automatized offline performance measurements. An evaluation showed that our technology achieves a prediction accuracy close to 95% and prediction evaluation times of a few milliseconds.

A configuration of an information technology/network (IT/NW) service is composed of components (e.g., applications, servers, and switches), relationships among them, and their attributes. To run a service appropriately, its configuration should be carefully organized to satisfy customer requirements, as well as dependencies and constraints derived from the components. Designing such a service configuration is daunting when the service is deployed in a heterogeneous environment including multiple clouds, edge devices, and different types of existing servers and network nodes. Thus, the time and cost for the designing task are serious problems in providing new services. In this paper, we present Weaver, an automated service configuration designer that generates a concrete service configuration on the basis of abstract customer requirements and the environment in which the service is deployed. Weaver accepts information about such requirements and environment as input and converts it into a fully concretized service configuration that can be deployed in the designated environment. In this paper, we take an example of a video surveillance Internet of Things (IoT) service, where multiple components including cameras, video analyzers and terminals, are distributed over multiple locations. In the evaluation, we show that Weaver produces configurations for designated heterogeneous environments. We also show that design time is dramatically shorter than in traditional system integration procedures. For example, the design time is shorter than a minute when the number of components is less than 150.

IoT(Internet of Things), which collects various data in real world and analyzes values from collected data is one of good methods to help solve such serious problems and to construct efficient social systems. Meanwhile, since collected data is very complicated and has huge size, it takes a long time to collect and analyze “Complicated Big data”. Then, efficient computer systems and efficient network systems are necessary. Integration of heterogeneous computing and 5G network is one of the best platforms to provide complex IoT systems and services. In this paper, first, a reason why complex IoT systems require high performance hetero computing and high-speed communication systems like as 5G is presented. In the next, some use cases of IoT systems infrastructures empowered by hetero computing are also introduced.

We have proposed an autonomous decentralized control using a local action rule for indirectly controlling the probability distribution of a system performance variable on the basis of markov chain monte carlo, while not measuring the variable. In this paper, we design an autonomous decentralized adaptive function for retaining the control strength of our control under a changing environment as an example of global controls appropriately reflecting information of external environment. We apply our control with the adaptive function to a virtual machine placement problem in a Data Center Network (DCN). Through simulation experiments, we confirm that the adaptive function effectively deals with several scenarios with a changing environment in a DCN.

Flow-based networks such as OpenFlow-based networks have difficulty handling a large number of flows in a node due to the capacity limitation of search engine devices such as ternary content-addressable memory (TCAM). One typical solution of this problem would be to use MPLS-like tunneling, but this approach spoils the advantage of flow-by-flow path selection for load-balancing or QoS. We demonstrate a method named "Source Flow" that allows us to handle a huge amount of flows without changing the granularity of flows. By using our method, expensive and power consuming search engine devices can be removed from the core nodes, and the network can grow pretty scalable. In our demo, we construct a small network that consists of small number of OpenFlow switches, a single OpenFlow controller, and end-hosts. The hosts generate more than one million flows simultaneously and the flows are controlled on a per-flow-basis. All active flows are monitored and visualized on a user interface and the user interface allows audiences to confirm if our method is feasible and deployable.

We propose a programmable and scalable traffic management scheme. Programmable traffic management at high-speed routers is difficult because programmability and high-speed packet processing have involved a serious tradeoff. To attain both, the new scheme combines control programs at a control server and simple packet handling functions, such as sampling packet headers and discarding packets, at routers. Therefore, by installing appropriate control programs into the server, a variety of active queue management schemes, per-flow bandwidth management schemes, DoS mitigation schemes, and so on, are achieved. One of the main contributions of this paper is its proposal of a statistical scheme for handling flows. As only a fraction of complete flow information stored at the control server is loaded into the router's flow table and it is replaced cyclically, the proposed scheme scales more than the router's flow table capacity. Our simulation results indicate that the scheme provides efficient traffic management, per-flow WFQ emulation in our example, even with very small flow tables compared to the number of concurrently active flows. Furthermore, we discuss implementation issues with the proposed scheme and reveal that the processing cost at the server and router is sufficiently small for use with 10 Gbps links.

In this paper, we propose a simplified multi-path aggregation scheme, Ethernet with flow label for multi-path, called EFL-MP. There are several aggregation schemes, such as link aggregation and multi-path TCP, for multi-path load balancing. However, link aggregation does not distribute a single flow into multiple paths due to its MAC address based flow dispatch. In addition, the multipath TCP scheme is not suitable for implementing into hardware because of the implementation complexity caused by the management of the two stages of sequence number (SEQ) (i.e., the SEQ assigned to a path and the SEQ assigned to the flow); thus it fails to provide high throughput and short latency. To maximize data transfer throughput and minimize latency, we employed a single-stage sequence number scheme with a reorder-buffer-usage-based retransmission activating algorithm. Simulation results show that EFL-MP provides high throughput and short latency despite its simple scheme, and its throughput and its latency suit to use for interconnection transport. In addition, we demonstrate that our proposed retransmission activating algorithm activates retransmission due to packet losses earlier than the conventional timer-based retransmission activating algorithm.

In this paper, we present the design, implementation and evaluation of a new TCP protocol, TCP-AW (TCP adaptive Westwood). This study was motivated by the intent to address simultaneously several challenging network scenarios, including high bandwidth efficiency in long and fat pipes, RTT fairness, and friendliness to legacy protocol, in a single protocol. TCP-AW leverages the key features of TCP Westwood and TCP adaptive Reno, namely, eligible rate estimation and delay- based adaptive AIMD parameter tuning, respectively. Extensive simulation and measurement results show that TCP AW yields good utilization of available bandwidth and achieves better RTT fairness. As for coexistence with TCP-NewReno, our major safety objective, was that new protocol compares more favorably than other promising protocols such as Hamilton-TCP, CUBIC and Compound-TCP. In contrast to other proposals, TCP-AW has no notions of small and large bandwidth delay product, instead, it scale seamlessly from current Internet paths to faster long distance paths.

Video streaming, including VOD (Video on Demand) services, over the Internet has been rapidly becoming popular. Although RTP/UDP has been considered as a standard transport protocol for video streaming, TCP is already widely used for VOD services because of its flexible accessibility to user clients beyond firewalls or NATs. However, poor video quality, for example, frequent pause of playback, due to TCP congestion control has been pointed out. In this paper, we propose an enhancement for TCP congestion control algorithm, which we call TCP-AV, to realize stable video streaming using TCP. TCP-AV incorporates two principal mechanisms, (i) dynamic TCP parameter tuning to stabilize TCP throughput around the target rate, and (ii) temporal target rate reduction to avoid severe congestion. Simulation results show that TCP-AV provides better rate control for maintaining target rate, and thus better video quality, even when the network is shared by many co- existing flows. In our scenario for a metropolitan VOD service, TCP-AV accommodates roughly 3 times larger number of VOD flows compared to TCP-Reno.

Multi-homed environments are increasingly common, especially for mobile users. To efficiently utilize multiple access lines for single file transfer, multipath TCP communication methods have been proposed. A multipath TCP enables simultaneous distributed data transfer between two end-points on multiple TCP connections. However, these methods cannot fully utilize the available bandwidth of multiple paths because they do not properly consider the end-to-end delay of packet transmission, so out-of-order data arrival at a receiver causes a bottleneck in data sort operations. This problem is more severe in environments where the quality of each path is different or unstable, such as in wireless environments. To solve this problem, we propose a multipath TCP communication method that includes a data distribution method to enable in-order delivery at a receiver. We call this arrival-time matching load-balancing (ATLB). ATLB continuously calculates the delay of each path, including the TCP queuing delay at a sender and the network delay, and then sends a data segment through the TCP connection with the lowest end-to-end delay. Simulation results show that ATLB improves end-to-end throughput, especially in heterogeneous environments where the quality of paths differs. For example, ATLB enabled twice the throughput with the conventional multipath TCP. We also report performance evaluation results from our ATLB test bed system in a wireless network environment. Our ATLB test bed system was able to fully utilize the aggregate available bandwidth of unstable multiple wireless links.

It has been recognized that current TCP (mostly TCP-Reno) throughput deteriorates in high-speed networks with large bandwidth-delay-product. A number of protocols, such as high speed TCP and scalable TCP, have been proposed to address this problem. However, their lack of friendliness to existing protocols has hampered their wide deployment in public networks. In this paper, we propose TCP-AR (adaptive Reno) to ensure friendliness to TCP-Reno, as well as efficiency in high-speed networks. A key feature of TCP-AR is that it dynamically adjusts the TCP response function based on congestion level estimation via RTT measurement. Namely, it increases congestion window faster and decreases the window less than TCP-Reno when it recognizes no congestion. As the congestion level increases, it tunes the response function so that it behaves like TCP-Reno. Simulation results show that TCP-AR maintains friendliness to TCP-Reno in networks with varying buffer capacities with or without RED, and flows with varying RTTs, while it achieves much higher throughput than TCP-Reno or even high speed TCP.

The appearance of various quality-sensitive applications has greatly changed the requirements for network management. To manage the quality of these applications, monitoring of individual traffic flows, as well as aggregated traffic statistics, has become more important. Since per-flow monitoring involves a high processing cost, especially in emerging high-speed links, packet sampling techniques have been attracting considerable attention. However, existing sampling techniques, such as NetFlow and sFlow, have mainly targeted traffic volume monitoring and there has been little discussion on the monitoring of quality indexes including packet loss rate. In this paper, we propose a method to estimate the TCP packet loss rate from sampled packets. The proposed method detects packet loss events by monitoring duplicate ACK events induced by a TCP receiver indicating the loss events. Since only a portion of packet loss events can be detected from the sampled packets, the correct packet loss rate is estimated by means of statistical approximation. Simulation results show that the proposed method accurately estimates the TCP packet loss rate from 10% of sampled packets.

In this paper, we propose a new version of TCP to improve (1) efficiency in wired-wireless combined networks with nonnegligible random packet losses, and also (2) friendliness to existing protocols, such as TCP-Reno. TCP-Westwood (TCPW) was proposed to improve efficiency in such networks; however, it is shown to be unfriendly to existing protocols under certain RTT and/or router buffer capacities. Since friendliness to existing protocols is one of the most important issues in a real network environment where different protocols coexist, we propose TCPW-BBE (TCPW with buffer and bandwidth estimation) to ensure the friendliness even under varying effective buffer capacities. Based on buffer capacity estimation mechanism, TCPW-BBE reacts more appropriately to a packet loss event, whether the loss is due to congestion or link errors. Simulation results show that TCPW-BBE maintains friendliness to TCP-Reno in networks for a broad range of buffer capacities, RTT, with/without RED routers, yet retaining the efficiency of the original TCPW.

An end-to-end "foreground/background" priority scheme is useful for end hosts to utilize the residual capacity left unused by high-priority foreground applications. Several end-to-end prioritization schemes, such as TCP-LP (Low Priority) and TCP-Nice, have been proposed, however, the residual capacity cannot be fully utilized by these schemes. We propose TCP Westwood Low Priority (TCPW-LP), a scheme that maximizes the utilization of residual capacity without intrusion on coexisting foreground flows. TCPW-LP employs an "Early Window Reduction" mechanism to reduce its congestion window as a reaction to incipient congestion. To achieve high efficiency, the reaction is based on the estimation whether the congestion is caused by the foreground traffic or not. Simulation results show that TCPW-LP appropriately defers to foreground flows. Further, under a wide range of buffer capacity and link error losses, TCPW-LP better utilizes the residual capacity than other proposed priority schemes or even TCP Reno.

Because of the development of recent broadband access technologies, fair services among users are becoming more important criteria. The most promising scheme of router mechanisms for providing fair service is per-flow traffic management. However, it is difficult to be implemented in high speed core routers because per-flow state management is prohibitive; thus, a large number of flows are aggregated into a small number of queues. This is not a preferable situation because the more number of flows aggregated into a queue increases, the worse fairness tends to become. In this paper, we propose a new traffic management scheme called hierarchically aggregated fair queueing (HAFQ) to provide per-flow fair service. Our proposed scheme can adjust flow aggregation levels according to the queue handling capability of various routers. That means the proposed scheme is scalably used in high-speed networks. HAFQ improves the fairness among aggregated flows by estimating the number of flows aggregated in a queue and allocating bandwidth to the queue proportionally. In addition, since HAFQ can identify flows having higher arrival rates simultaneously in estimating the number of flows, it enhances the fairness by preferentially dropping their packets. We show that our proposed scheme can provide per-flow fair service through extensive simulation and experimental studies using a network processor. Since the currently available network processors (Intel IXP1200 in our case) are not high capacity, we also give extensive discussions on the applicability of our scheme to the high-speed core routers.

The assured forwarding per hop behavior standardized by the IETF Differentiated Services working group provides four class-based differentiated IP services. In this service, however, unexpected service degradation may occur and differentiation among classes may be disordered if the network is designed to minimize over-provisioning or is under-provisioned. We therefore developed a packet scheduling scheme that dynamically allocates bandwidth to each class queue to guarantee the differentiation among classes under any traffic conditions. The scheme estimates the sum of CIRs (committed information rates), i.e. rate of the packets having lowest drop preference, of active flows in each class and initially allocates the link bandwidth according to the sum of CIRs. It allocates the excess bandwidth by using a combination of CIR-proportional allocation and equal-share allocation. The equal share part enables that the flows in best effort class or the flows having zero CIRs can utilize minimum share of the bandwidth. Our scheme also introduces a scalable scheduling technique to improve fairness among flows in the same class. We evaluate the proposed scheme and show that it makes DiffServ operations fairer under any traffic conditions.

The FRP (fast reservation protocol) utilizes a unique feature of ATM (asynchronous transfer mode) technology. The FRP is supposed mainly to be applied to LAN interconnection. We provide an exact analysis for a class of the FRP to obtain the burst (a protocol data unit of FRP) level performance of the FRP. One of the main features of the ATM is that the transmission rate can be adjusted according to the network congestion states. In our modeling, this feature is incorporated in such a way that the bandwidth a source requests is reduced if an attempt to reserve the bandwidth is rejected by the network. A rationale behind this is that rejection of the bandwidth request indicates network congestion. Therefore, the request with the smaller bandwidth after the reservation failure enables better sharing of network resources, which results in a performance improvement. To solve our model, which contains a very huge number of system states, we introduce a new numerical approach which is an extension of Sumita and Rieders's (1991) replacement process approach. Through numerical examples, the appropriate transmission rate control method is examined, and considerations on backoff time and overhead of RM cells are also presented.