Zhen Wan, Chao-Han Huck Yang, Jinchuan Tian, Hanrong Ye, Ankita Pasad, Szu-Wei Fu, Arushi Goel, Ryo Hachiuma, Shizhe Diao, Kunal Dhawan, Sreyan Ghosh, Yusuke Hirota, Zhehuai Chen, Rafael Valle, Chenhui Chu, Shinji Watanabe, Boris Ginsburg, Yu-Chiang Frank Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Practical optical buffers have not yet been implemented in optical packet switching (OPS) networks, where packet-by-packet switching is performed using optical signals at optical switches. As a result, packet collisions occur when multiple optical packets simultaneously reach the same output port of optical switches. This leads to significant communication performance degradation, particularly when Transmission Control Protocol (TCP) is used. In this paper, we propose a method that the receiver host adaptively delays the transmission of TCP acknowledgment (ACK) to achieve a predefined target round-trip time (RTT) for the sender host. This target RTT is determined based on real-time pure RTT measurements obtained during data transmission and reception. By dynamically adjusting the RTT of data packets traversing optical paths of various lengths in an OPS network to one of a limited set of fixed values, we can reduce the likelihood of such collisions.

In an Optical Packet Switching (OPS) network environment, the absence of practical optical buffers results in frequent packet collisions at intermediate nodes, as multiple packets may arrive simultaneously at the same output port of a certain intermediate node. This leads to substantial packet loss even under non-congested conditions. Consequently, the conventional Transmission Control Protocol (TCP) does not fully utilize the bandwidth of the bottleneck links. Moreover, highly bursty data transmission can impose the above impact excessively on intermediate nodes in OPS networks, making it undesirable. To address this issue, we have presented a control mechanism consisting of two disciplines: microburst transmission and congestion window updating. However, the performance of microburst transmission control itself, which corresponds to TCP pacing, has not been quantitatively clarified. In this paper, we quantitatively evaluate the performance of microburst transmission control. Specifically, we investigate the impact of different microburst transmission intervals on throughput and clarify the potential fairness of throughput among optical paths with different route lengths.

Gender bias in vision-language foundation models (VLMs) raises concerns about their safe deployment and is typically evaluated using benchmarks with gender annotations on real-world images. However, as these benchmarks often contain spurious correlations between gender and non-gender features, such as objects and backgrounds, we identify a critical oversight in gender bias evaluation: Do spurious features distort gender bias evaluation? To address this question, we systematically perturb non-gender features across four widely used benchmarks (COCO-gender, FACET, MIAP, and PHASE) and various VLMs to quantify their impact on bias evaluation. Our findings reveal that even minimal perturbations, such as masking just 10% of objects or weakly blurring backgrounds, can dramatically alter bias scores, shifting metrics by up to 175% in generative VLMs and 43% in CLIP variants. This suggests that current bias evaluations often reflect model responses to spurious features rather than gender bias, undermining their reliability. Since creating spurious feature-free benchmarks is fundamentally challenging, we recommend reporting bias metrics alongside feature-sensitivity measurements to enable a more reliable bias assessment.

Large-scale vision-language models, such as CLIP, are known to contain societal bias regarding protected attributes (e.g., gender, age). This paper aims to address the problems of societal bias in CLIP. Although previous studies have proposed to debias societal bias through adversarial learning or test-time projecting, our comprehensive study of these works identifies two critical limitations: 1) loss of attribute information when it is explicitly disclosed in the input and 2) use of the attribute annotations during debiasing process. To mitigate societal bias in CLIP and overcome these limitations simultaneously, we introduce a simple-yet-effective debiasing method called SANER (societal attribute neutralizer) that eliminates attribute information from CLIP text features only of attribute-neutral descriptions. Experimental results show that SANER, which does not require attribute annotations and preserves original information for attribute-specific descriptions, demonstrates superior debiasing ability than the existing methods.

The networking industry is offering new services leveraging recent technological advances in connectivity, storage, and computing such as mobile communications and edge computing. In this regard, extended reality, a term encompassing virtual reality, augmented reality, and mixed reality, can provide unprecedented user experience and pioneering service opportunities such as: live concerts, sports, and other events; interactive gaming and entertainment; immersive education, training, and demos. These services require high-bandwidth, low-latency, and reliable connections, and are supported by next-generation ultra-reliable and low-latency communications in the vision of 6G mobile communication systems. In this work, we devise a novel scheme, called backup from different data centers with multicast and adaptive bandwidth provisioning, to admit reliable, low-latency, and high-bandwidth extended reality live streams in next-generation networks. We consider network services where contents are non-cacheable and investigate how backup services can be offered by different data centers with multicast and adaptive bandwidth provisioning. Our proposed service-provisioning scheme provides protection not only against link failures in the physical network but also against computing and storage failures in data centers. We develop scalable algorithms for the service-provisioning scheme and evaluate their performance on various complex network instances in a dynamic environment. Numerical results show that, compared to conventional service-provisioning schemes such as those seeking backup services from the same data center, our proposed service-provisioning scheme efficiently utilizes network resources, ensures higher reliability, and guarantees low latency; hence, it is highly suitable for extended reality live streams.

We investigate the impact of deep generative models on potential social biases in upcoming computer vision mod-els. As the internet witnesses an increasing influx of AI-generated images, concerns arise regarding inherent biases that may accompany them, potentially leading to the dis-semination of harmful content. This paper explores whether a detrimental feedback loop, resulting in bias amplification, would occur if generated images were used as the training data for future models. We conduct simulations by progressively substituting original images in COCO and CC3M datasets with images generated through Stable Dif-fusion. The modified datasets are used to train OpenCLIP and image captioning models, which we evaluate in terms of quality and bias. Contrary to expectations, our findings indicate that introducing generated images during training does not uniformly amplify bias. Instead, instances of bias mitigation across specific tasks are observed. We further explore the factors that may influence these phenomena, such as artifacts in image generation (e.g., blurry faces) or pre-existing biases in the original datasets.

Yusuke Hirota, Jerone Andrews, Dora Zhao, Orestis Papakyriakopoulos, Apostolos Modas, Yuta Nakashima, Alice Xiang. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Large language models (LLMs) have enhanced the capacity of vision-language models to caption visual text.This generative approach to image caption enrichment further makes textual captions more descriptive, improving alignment with the visual context.However, while many studies focus on the benefits of generative caption enrichment (GCE), are there any negative side effects?We compare standardformat captions and recent GCE processes from the perspectives of "gender bias" and "hallucination", showing that enriched captions suffer from increased gender bias and hallucination.Furthermore, models trained on these enriched captions amplify gender bias by an average of 30.9% and increase hallucination by 59.5%.This study serves as a caution against the trend of making captions more descriptive.Code is available at https://github.com/rebnej/ GenCapEnrich-bias.

Summary We consider automatic performance tuning of dense symmetric eigenvalue problems using ATMathCoreLib, which is a library to assist automatic tuning. We deal with two problems, namely, automatic code selection for the symmetric generalized eigenvalue problem in distributed‐memory parallel environments and automatic parameter tuning in tridiagonalization of dense symmetric matrices on multicore processors. As for the first problem, numerical experiments show that ATMathCoreLib can choose the fastest solver for a given computing environment and problem size quickly even if the fluctuation in the execution time is as high as 40%. As for the second problem, ATMathCoreLib was able to select nearly optimal combinations of the algorithm and its parameter reliably and efficiently for various computing environments and matrix sizes. The performance of auto‐tuning was further enhanced by incorporating a user‐provided execution‐time model into ATMathCoreLib.

In lossy networks, including optical packet switching (OPS) networks and wireless networks, data recovery from frequent packet loss is essential. Erasure-coding, a type of forward error correction (FEC), is one of effective solutions, and furthermore, it could be combined with a technique that feeds back decoding progress information from the receiver to the sender for efficient data transmission. However, conventional feedback methods cannot inform the sender of the decoding progress information at the receiver when acknowledgment (ACK) packets are lost. Therefore, we propose a new data transmission method that is robust to high packet loss by feeding back accumulated information of decoded symbols at the receiver, and furthermore that smartly reduces the calculation overhead of encoding/decoding process. The proposed method can reduce the redundancy by up to 5.84% compared to conventional methods in environments where random packet loss occurs.

The increasing tendency to collect large and uncurated datasets to train vision-and-language models has raised concerns about fair representations. It is known that even small but manually annotated datasets, such as MSCOCO, are affected by societal bias. This problem, far from being solved, may be getting worse with data crawled from the Internet without much control. In addition, the lack of tools to analyze societal bias in big collections of images makes addressing the problem extremely challenging. Our first contribution is to annotate part of the Google Conceptual Captions dataset, widely used for training vision-and-language models, with four demographic and two contextual attributes. Our second contribution is to conduct a comprehensive analysis of the annotations, focusing on how different demographic groups are represented. Our last contribution lies in evaluating three prevailing vision-and-language tasks: image captioning, text-image CLIP embeddings, and text-to-image generation, showing that societal bias is a persistent problem in all of them. https://github.com/noagarcia/phase

Image captioning models are known to perpetuate and amplify harmful societal bias in the training set. In this work, we aim to mitigate such gender bias in image captioning models. While prior work has addressed this problem by forcing models to focus on people to reduce gender mis-classification, it conversely generates gender-stereotypical words at the expense of predicting the correct gender. From this observation, we hypothesize that there are two types of gender bias affecting image captioning models: 1) bias that exploits context to predict gender, and 2) bias in the probability of generating certain (often stereotypical) words because of gender. To mitigate both types of gender biases, we propose a framework, called LIBRA, that learns from synthetically biased samples to decrease both types of biases, correcting gender misclassification and changing gender-stereotypical words to more neutral ones.

With the adoption of edge computing, several data centers are available within the footprint of an optical metro network, and contents are replicated in multiple locations. Such a wide content replication offers a unique opportunity to provide better services to users, especially for content-based services, e.g., video delivery. Thus, a service-provisioning scheme can embrace this opportunity to optimize network resource utilization, improve reliability, and achieve lower latency. In this study, we propose a reliable service-provisioning scheme that selects the optimal subset of data centers hosting the desired content and inversely multiplexes a content request over multiple link-disjoint paths. We formulate an integer linear program and develop heuristics for the problem, and use them to solve various complex and realistic network instances. Numerical data show that, compared to conventional service-provisioning schemes such as multipath routing from a single data center or dedicated-path protection, our proposed scheme efficiently utilizes network resources, improves reliability, and reduces latency; hence, it is suitable for the above-mentioned services.

With the evolution of optical networks, it is expected that each server and user terminal will be equipped with an optical interface and directly connected to all-optical networks. In particular, Optical Packet Switching (OPS) network is one of the candidate future networks, where it can realize ultra-high-speed transmission with a highly statistical multiplexing effect. Thus, in addition to simple one-to-one high-speed connections, Service Function Chaining (SFC), including the selection of servers, optical route, and signal resources, needs to be provided on an optical network operation framework. In this paper, we demonstrate the possibility of selecting appropriate servers and optical signal-level communication resources with an optical network testbed. As a realistic solution to realize SFC on an OPS network operation, we implement a centrally controlled OPS network with software-defined network technology. We verify the dynamic selection of servers and optical communication resources based on aggregated load information of servers and confirm the feasibility of our SFC management.

We study societal bias amplification in image captioning. Image captioning models have been shown to perpetuate gender and racial biases, however, metrics to measure, quantify, and evaluate the societal bias in captions are not yet standardized. We provide a comprehensive study on the strengths and limitations of each metric, and propose LIC, a metric to study captioning bias amplification. We argue that, for image captioning, it is not enough to focus on the correct prediction of the protected attribute, and the whole context should be taken into account. We conduct extensive evaluation on traditional and state-of-the-art image captioning models, and surprisingly find that, by only focusing on the protected attribute prediction, bias mitigation models are unexpectedly amplifying bias.

Cooperation among datacenter providers (DCPs) and network carriers is necessary to support today's ubiquitous cloud services. However, such cooperation can be constrained by limited visibility as confidential information, such as network topology, resource availability, etc., may not be disclosed among these entities due to regulatory policies. We study a DCP-carrier cooperation-based service restoration scheme during a disaster with the aid of a third-party mediator, namely a Provider Neutral Exchange (PNE). We propose a novel resource-driven demand-matching strategy to restore DCP services. When multiple DCPs compete for network resources (due to post-disaster resource crunch), resource balancing by PNE can achieve fair and efficient service restoration. To allow flexibility in demand-resource matching, DCPs generate multiple sets of connection requests and define varying priorities and bandwidth degradations for each request. Carriers evaluate the DCP requests and provide feedback (e.g., whether a request can be satisfied or not) based on their available resources. We present an eight-phase DCP-carrier cooperation framework, with each phase employing individual sub-tasks carried out by DCPs, carriers, and PNE. Results under different disaster scenarios show that our strategy significantly improves DCP service restoration, incurring less restoration time.

Large-scale disasters affecting both network and datacenter (DC) infrastructures can cause severe disruptions in cloud-based services. During post-disaster recovery, repairs are usually carried out in stages in a progressive manner due to limited repair resource availability. The order in which network elements and DCs are repaired can significantly impact users' reachability to important contents/services. We investigate joint progressive network and DC recovery in which network recovery and DC recovery are conducted in a coordinated manner such that users have access to the maximum possible amount of contents/services at each repair stage. We first solve the optimization problem of joint progressive recovery to find the optimal sequence of network element and DC repairs with the objective to maximize cumulative weighted content reachability in the network. We then propose a scalable heuristic for scheduling the sequential repair of network nodes/links and DCs. Our model assumes that, at each repair stage, one network node with adjacent links and one DC can be fully repaired; however, full recovery may not be guaranteed due to limited resource availability. Hence, we also propose a “resource-aware” approach (with two resource-allocation strategies, namely “selective allocation” and “adaptive allocation”), which considers both full and partial recovery of elements based on available resources at each stage. We show that, compared to disjoint progressive recovery approach, in which network recovery and DC recovery plans are independent, our joint progressive recovery approach provides significantly higher per-stage content reachability in the network.

Multi-homed wireless video streaming systems concurrently send video frames of one video content from multiple wireless access networks. The coexistence of Long Term Evolution (LTE) and Wi-Fi systems is one of the typical multi-homed wireless environments. To utilize such multiple wireless access links for high-quality streaming, we need to overcome two issues: rate decrement in the LTE channel due to cellular congestion and channel quality fluctuation in Wi-Fi channels due to noise, interference, and channel fading. To this end, we propose a hybrid digital-analog video transmission scheme, which integrates digital video coding, near-analog modulation, and compressive sensing, considering Wi-Fi offloading. To avoid an effect of the cellular congestion, the proposed scheme offloads a part of digital-encoded video data, which are originally transmitted in the LTE channel, to available Wi-Fi channels if LTE's data rate becomes insufficient during the streaming session. In addition, the proposed scheme uses a combination of compressive sensing and near-analog modulation for video delivery in Wi-Fi channels to ensure the received video quality can be proportionally improved according to the instantaneous quality of the Wi-Fi channels. From evaluations, the proposed Wi-Fi offloading technique keeps better video quality even when the capacity of LTE channel becomes limited within a streaming session. In addition, it was found that the received video quality can be gracefully improved according to the number of receivable Wi-Fi access points and each Wi-Fi channel quality.

Recent progress of computer networks realizes advanced network services such as high-resolution video streaming and Internet of things (IoT) but it requires further huge amount of traffic. To expand transmission capacity, various studies develop time division multiplexing (TDM), wavelength division multiplexing (WDM), and spatial division multiplexing (SDM) technologies. Time-division packet spatial super-channel (pSSC) is expected as one of promising solutions to realize high throughput with low cost. We have developed a switching node system for the pSSC transmissions. The pSSC is suitable for large data transmission. On the other hand, overhead of pSSC is not negligible in the case of small size data transmissions. This paper proposes a fractional pSSC switching system design for both large and small data transmissions. We introduce the spatial super-channel slicing concept and ILP solution for effective data transmission having bimodal distribution of data size. Effect on slice spatial channel was validated through simulation where it is 50% better than no slicing system.

In conventional multi-view video systems using progressive download, each user downloads only a desired viewpoint for traffic reduction; however, playback stalls occur after every view switching. The playback stalls induce low user satisfaction for such multi-view applications. A progressive download-based multi-view video scheme assuming successive view switching has been proposed to reduce video traffic and prevent playback stalls. However, when the user switches to a viewpoint other than adjacent viewpoints, this still causes playback stalls when watching the desired viewpoint. In particular, when a user randomly switches an observed viewpoint to find the preferred viewpoint, i.e., zapping, it needs to send videos of all viewpoints to accept zapping. This also causes playback stalls owing to sharp traffic growth. In this paper, we propose a novel progressive download-based multi-view video delivery scheme to decrease the number of playback stalls even when the user switches to an observed viewpoint other than adjacent viewpoints, including zapping. The main idea of the proposed scheme is to execute segment download scheduling by taking into account the view switching probability and viewpoint popularity of past users. Specifically, a user node foresees a user's view switching behavior in future video frames based on those information and schedules a download order of priority segments, which will be highly likely to be switched by the user, for smooth view switching. Evaluations using Joint Multi-view Video Coding (JMVC) encoder and multiview video sequences show that our scheme decreases the number of playback stalls.

With the development of communication services, backbone networks need to deal with various types of traffic by utilizing appropriate switching systems in a flexible manner. Optical packet and circuit integrated (OPCI) networks offer both optical packet-switching (OPS) and optical circuit-switching (OCS) and have been studied for this purpose. A space division multiplexing-based OPCI network provides significant communication capacity and shares spectrum resources for OPS and OCS traffic demands on the same physical infrastructure to efficiently accommodate future services. For appropriate resource utilization, the ratio of spectrum resources between OPS and OCS must be flexibly managed based on the dynamically changing traffic. One of important issues of OPCI networks is sharing spectrum resources under a dynamic traffic model. In this paper, we propose a dynamic spectrum-sharing method for both OPS and OCS data transmissions to enhance resource utilization. The spectrum-sharing method consists of dynamic resource boundary management and load-balancing by exploiting integer linear programming periodically in operation. We evaluate the proposed method with computer simulation and verify its effectiveness in terms of both OPS loss probability and OCS blocking probability.

Wireless video delivery using multiple transmission paths has been studied in recent years to improve the received video quality. Hybrid WiFi-VLC is one of the promising multi- path techniques for indoor communications. Hybrid WiFi- VLC uses radio and visible light bands for transmissions. When a sender simply transmits videos over the hybrid path, the received video quality will be low due to two features of the visible light band. First one is the quality of visible light path will be significantly changed depending on the position and angle of devices, unlike the conventional radio communication. The second one is visible light path is typically one-way link. It means the conventional schemes have not been possible to provide video contents of adequate quality for each user. In order to address the above-mentioned issues, this paper proposes a video transmission scheme for hybrid WiFi-VLC environments. The proposed scheme integrates two-dimensional discrete wavelet transform (2D-DWT), digital video coding, and pseudo-analog modulation to provide baseline quality over radio paths and realize graceful quality enhancement according to the quality of visible light paths without the channel state information. Evaluations show that the proposed method improves PSNR performance by 11 dB compared to a simple method of dividing and transmitting video data over the radio and visible light paths across signal-to-noise ratios of 0 to 25 dB.

Real-time multi-view video streaming plays an important role in new interactive and augmented video applications such as telepresence, remote surgery, and entertainment. Multiview video streaming approaches exploit time and inter-camera domain correlations in video frames for compression, thereby obtaining two benefits: video traffic reduction and video quality maintenance. However, excessive video encoding requires long periods of time for encoding/decoding computation in real-time multi-view video streaming and this process is vulnerable to frame loss, both of which may induce stalling and skipping during video playback at the user node. To address these problems, we propose Adaptive Multi-view Video Streaming (AMVS) system for real-time streaming. AMVS employs a greedy algorithm to adaptively select a prediction structure with low encoding/decoding computation based on the available network bandwidth. AMVS encodes and transmits multi-view video by considering the similarity of the user's request and the efficiency of transmission. Evaluations using Joint Multiview Video Coding (JMVC) demonstrated that AMVS achieves low encoding/decoding computation and high video quality while satisfying bandwidth limitations.

Various video distribution services have emerged, such as Ultra High Definition (UHD) videos, 3D videos and multi-view videos. Despite the fact that the issue of video distribution in best-effort networks is addressed in vast amount of literatures, little work has been done so far to consider video distribution systems in bandwidth guaranteed networks. In this paper, we propose a P2P-based UHD multi-view video distribution system in order to realize high quality play-back and rapid view switch by distributing UHD multi-view video in best-effort and guaranteed networks cooperatively. Through computer simulations, we demonstrate that the proposed system is suitable for future UHD multi-view video distribution systems.

Progressive download for multi-view video delivery is one of promising techniques to provide immersive and interactive experiences to users. In conventional multi-view video systems, a user downloads videos of all viewpoints of one content to realize smooth view switching. However, it causes increase of video traffic, and thus low video quality within an available download rate. To download only a desired viewpoint is another approach for reduction on volume of video traffic. However, playback stalls occur after view switching. The stalls induce low user's satisfaction for applications. In this paper, we aim at two objectives: 1) to achieve reduction on video traffic and 2) to achieve reduction on the number of playback stalls. To this end, we propose a new multi-view video delivery scheme for progressive download. The main idea of the proposed scheme is that the user only downloads potential videos, which are potentially played back by the user, to realize both traffic reduction and smooth view switching. In addition, we propose two download scheduling algorithms to prevent playback stalls even in a low download rate. The first algorithm prevents stalls in frequent view switching cases while the other prevents stalls in gazing cases. Evaluations using Joint Multi-view Video Coding (JMVC) encoder and multi-view video sequences show that the proposed method reduces video traffic by 55.1% compared with a simulcast scheme. In addition, the proposed method decreases the number of playback stalls during video playback by 86.0% compared with a request and response scheme.

Optical circuit and packet integrated network is a promising network architecture that can deal with various sizes/types of data efficiently. In this integrated network, paths (routes) for circuit and packet switching are assumed to be calculated and selected independently. However, this approach is insufficient in terms of efficiency and rapidity of total path calculation. Calculating appropriate paths with shorter computation time is necessary. In this paper, we propose and evaluate a high-speed multipath computation method that quickly calculates end-to-end paths for optical circuit switching by making use of the calculation results of next hop information for packet switching.

The multi-view video streaming is one of promising technologies for emerging video services. For those applications of multi-view video streaming, video frames of all viewpoints, i.e., cameras, are needed to be transmitted to viewers because the demands of all the viewers' view-switching are unpredictable. However, existing transmission schemes are highly vulnerable to frame loss. Specifically, the frame loss in one viewpoint induces a collapse of decoding for other viewpoint videos. To improve loss-resilience, we propose a multi-path based multi-view video transmission. Our scheme encodes video frames into multiple descriptions that are mutually independent of each other, by using inter-view prediction. It then transmits each description using multiple transmission paths. Our scheme makes three contributions:1) it reduces video traffic even for a large number of cameras, 2) it prevents an increase in the number of undecoded video frames caused by one frame loss, and 3) it conceals frame loss by using the video frames in the other paths. Our scheme generalizes to an arbitrary number of transmission paths and discusses the detailed performance at the number of the paths of 2. Evaluations show that our proposed scheme improves video quality by 3 dB compared to existing transmission schemes in loss-prone environments.

To guarantee the quality of service is one of the most important problems in ICT (Information Communication Technology) area. Therefore, we had proposed a service platform which integrates service control and network control by componentization of distributed objects. On the other hand, with advances of wireless communication technologies, users eager to have network services with a mobile terminal. However, this platform does not consider any wireless access networks. In this paper, we extend the platform to support mobile terminals with heterogeneous wireless access networks and also propose a traffic load balancing method on them for bandwidth guaranteed services. Finally, the effectiveness of the proposed method is evaluated by simulation experiments.

Video streaming using hierarchical cache servers is effective to provide high quality services to many users. However, less-accessed videos tend to be sparsely cached on cache servers far from users. Streaming qualities are degraded due to relatively high packet loss rate and long delay. In order to address the issues on heavy and bursty packet loss over such long distance transmission, we have focused on recovery characteristic of a multi-server Forward Error Correction (FEC) function named “Inter-Stream FEC.” In this paper, we propose a high quality streaming system exploiting hierarchical cache servers. The proposed system improves users' experience by adaptively applying Inter-Stream FEC. We also propose a parity data request method combines with parity server selection considering cache status of individual server and disjointness between paths. Simulation results show the proposed system can achieve higher restoration performance compared with a streaming system using traditional FEC, particularly when bursty loss occurs.

Recently, network services based on distributed components that provide services by using components running on remote nodes have attracted much attention. Advanced services can be provided at a lower cost by using existing components. We have previously proposed a method that ensures a stable QoS for each service by allocating components dynamically according to changes in access frequency. In that method, however, we considered only CPU resources rather than network resources. In this paper, we propose a method for providing bandwidth-guaranteed services stably by considering available wavelength resources in backend servers on a platform with a WDM optical network.

On real-time video streaming, the distribution with multi-tree Application Level Multicast (ALM) is effective in terms of delay and jitter. In constructing multi-tree networks, a joining node selects counterpart nodes based on their performance or bandwidth capacity and they are connected each other. However, tree-based ALM has the overhead of reconstructing the tree topology when any node leaves. Multi-tree ALM has the large overhead of reconstruction. In addition, ALM does not consider distance between nodes on IP networks. This causes increasing delay and wasting network resources. In this paper, for the purpose of churn resilience, reduction of traffic volume and short delay, we propose a construction method of overlapped cluster-trees considering inter-node distance for resilient live video streaming. The proposed construction method consists of two main functions; clustering neighbor nodes on IP network and adjusting the cluster size. The overlapped cluster-trees are resilient to churn. Clusters autonomously manage member nodes according to the cluster size. The performance of the proposed methods is evaluated through extensive computer simulation. The result shows that the proposed cluster-trees can decrease transmission delay and increase resilience to churn compared with conventional multi-tree ALM.

In this paper, we propose a video data recovery method that exploits multiple streams. The proposed method gives priority to each data of multiple streams in terms of recovery from packet loss, and it is designed for recovering important data with higher probability. The priority of each data is calculated based on both inherent importance in one stream and loss probabilities of multiple streams. Specifically, this method deals with the data of less received streams as the important data. Moreover, the proposed method decides appropriate number of parity packets. This is because the transmission of too many parity packets might affect the other traffic, while that of too few packets might recover insufficient amount of data. The proposed method can improve video quality of total multiple streams. Our simulation results show that the proposed method can recover more important data with higher probability. Furthermore, the proposed method can also achieve almost the same restoration ratio among streams of recovery targets.

Application level multicast with multiple distribution trees is effective for real-time streaming in terms of low delay and jitter. However, failure of upstream nodes drastically influences on the performance of many downstream nodes in multiple tree-based multicast. This degradation of video data continues until reconstruction of distribution trees finishes. Therefore, it is important to recover from the isolation as fast as possible. In this paper, we propose proactive and reactive approaches in order to resolve this problem. The proactive approach realizes fast recovery from isolation and efficient randomized forwarding by distribution trees cooperation. Each node forwards received data to its child nodes not only in the tree but also in other trees according to a pre-determined probability. In contrast, in the reactive approach, an isolated node requests forwarding from other nodes after detecting the parent node's trouble. This approach achieves quick response by requesting to near nodes. Moreover, we propose the hybrid method combining these two approaches. This hybrid method has advantages of both the proactive and reactive approaches. Finally, we evaluate the performances of the proposed methods through computer simulations, and confirm the effectiveness of the proposed methods.

In this paper, we propose a data recovery method that exploits multiple video streams. The proposed method gives priority to each data of multiple streams in terms of recovery from packet losses, and it is designed for recovering important data with higher probability. The priority of each data is set based on both inherent importance within a stream and loss probabilities of multiple streams. Specifically, this method deals with the data of less received streams as the important data. Moreover, we focus on the creation method of redundant data to be transferred from extra server for multiple worse conditioned streams, which greatly affects the recovery capability, and propose the creation method which is less influenced from the network condition. Our simulation results show that the proposed method can recover more important data with higher probability. Furthermore, the proposed method can also achieve almost the same restoration ratio among streams of recovery targets which have different loss probabilities.

Network services based on distributed components, which provide advanced network services with lower cost by flexibly assigning and reusing useful components running on remote servers, are attracting increasing attention. In this study, we propose a new platform which provides sustainable QoS and faster service time by allocating components dynamically with fault-tolerance. Specifically, we assume that many kinds of components are running on component servers and each component has its own Service Level Agreement (SLA). The proposed platform achieves to make a load-balancing system to meet SLA for service time. Moreover, the platform can provide faster service time to allocate components dynamically considering inter-component communication delay. Finally, to show the performance of our platform, we perform simulation experiments and prototyping.

P2P live streaming distribution services which can reduce load of distributor becomes commonly. The contents of these services are provided by not only corporations but also individuals. In existing systems for business services, however, transmission load of distributor still remains in high level to keep quality of service higher. Thus, it is difficult for private content distributors to provide for their own streaming contents. This paper presents and evaluates P2P live streaming system suitable for private distribution. This system constructs network considering transmission bandwidth of each node. In particular, a certain number of broadband nodes are carefully selected and set at the point nearby the distributor's node in P2P network. A set of the selected broadband nodes plays a role of a virtual distributor with large amount of transmission bandwidth and provides contents to peers instead of Origin. The original contribution of our proposal includes double mesh P2P network structure for scalable and efficient private live streaming services, and its detailed design description. In this method, network bandwidth is used effectively. As a result, it is expected to achieve the private distribution with higher scalability. With simulation experiments, we confirmed that the proposed system presents about half load of distributor compared with the existing system.

In this paper, we propose a new service platform which provides stable QoS by allocating components dynamically with considering fault-tolerance. We assume that many kinds of components are running on component servers and each component has its own SLA. The proposed platform achieves to make a load-balancing system to meet SLA of these components SLA and considering the number of replications necessary to offer a continuous service in case of failure. Finally, by implementation of its prototyping system, we confirm that the proposed platform is effective and feasible to provide the network services based on distributed components.