論文

マルチメディアシステムでは，新しいセッションを生成したり，連続メディア セッションが終了するときに返還される資源の再分配にセッション間の調停方 式が重要である．エンドユーザはセッションのサービスの質（Quality of Service; QOS）を何KBといった厳密なメモリサイズやネットワークバンド幅を 指定したくはないが，このような定量値はアドミッション制御には欠かせない． そこで，我々は3-階層化したQOS指定と層間のQOSの翻訳機構を導入してシステム が定量値を扱えるようにした．それでも，セッション間の調停には複雑で時に はNP-完全問題に相当する計算量が必要となる．そこで，QOS指定に単調性の制 約をつけることによって，セッション間調停を単純化し実装運用が可能になっ た．本論文では，Real-Time Mach 3.0上の Conductor/Performerシステムを用 いて，QOSの翻訳と複数セッション間調停の実装を報告する．

This paper presents the design and implementation of a user-level real-time network system in Real-Time Mach. Traditional network systems for microkernel based operating systems, which tend to focus on high performance and flexibility, are not suitable for real-time communication. Our network system provides a framework for implementing real-time network protocols which require to bound protocol processing time, and it is suitable for implementing on microkernel based operating systems. In this paper, we especially focus on the aspects to avoid the priority inversion problem in order to make network systems more preemptable and predictable. We also describe the feasibility of our network system for building distributed multimedia systems.

Quality of Service (QoS) control of continuous media such as video and audio with Operating System (OS) support has been extensively studied. On the network side, bandwidth-guaranteed networks are already available. Both the OS and the network are becoming able to support continuous media, but there has not been much discussion on collaboration between them for efficient handling of Variable Bit Rate (VBR) continuous media. We consider QoS control of VBR continous media in both end hosts, a sender and a receiver, and consider the delay of theApplication Data Unit (ADU). We propose ``Channel Redirection,

Java has many benefits such as security in distributed environments, reusability of code, and portability because it is architecture neutral. From those characteristics, Java is beginning to be used in many new environments. Even though Java provides various advantages, it still has problems which must be solved. One issue is that, there are often real-time constraints that should be met in those applications. Current Java execution environment and language specification cannot satisfy those requirements. Our research focuses on the current limits of the Java language and its execution environment, and to seek the possibility for real-time using Java and the technologies we have available today. To investigate issues in real-time Java, we have implemented a prototype real-time Java environment which provides real-time Java threads and real-time synchronization mechanism. Then we evaluated its performance. The results indicate that timing requirements are better met and unbounded priority inversion is avoided in our system.

In a distributed multimedia integration environment, there is a need for transmission control methods based on real-time communication protocols, that provide a relative Quality Of Service (QOS) to allow a variety of receiving capacities for group members and their requests and that provide a dynamic QOS to cope with temporary CPU overload and network congestion. Therefore, this study proposes a cooperative control method for end-to-end QOS control at a transport layer and for point-to-point control at a network layer. The former is mainly a set of flow-monitoring and flow-adjusting functions, and the latter a set of flow-control functions. The flow-adjusting functions provide in-depth evaluation of (1) the characteristics of the transmission data, (2) end-to-end network performance, and (3) the requirements of receiver applications. The appropriate QOS levels and flow control methods are then determined on the basis of the results.

Recent modern operating system technology enables protocol processing in user space using in-kernel packet filter and user-level protocol processing library for flexibility without sacrificing the performance of traditional kernelized protocol processing. This technology can be adapted to build a highly preemptable protocol processing mechanism for distributed real-time environment. In this paper, we discuss the structural difference of various operating systems from the protocol processing point of view, and propose an extended mechanism of packet filter and user-level protocol processing library for real-time communication. Using this mechanism, priority of the client can be handed off to the server without priority inversion problem while protocol processing.

電子通信情報学会誌に6回連載されたマルチメディアコンピューティングに関 する講座の第5回を担当し，ミドルウェアとツールキットに関して解説した．

次世代インターネットプロトコル IPv6 では、24 ビットのフローラベルや優 先度フィールドなど、実時間通信処理サービスへの要求に応じた QOS サポー トが考慮されており、今後の課題として(1)フローラベル毎のフロー制御、(2) ATM の QOS コネクションとのマッピング、(3)ルーティングプロトコルやNDP (Neighbor Discovery Protocol) とのキャッシュ情報の一貫性管理、(4)キュ ーでの優先度制御、(5)パスと資源予約制御が挙げられる。さらに、慶應大学 を中心とした次世代マイクロカーネル(MKng)研究プロジェクトにおける、RT- Mach (Realtime Mach) OS 上でのリアルタイムスレッドを用いた次世代インタ ーネットプロトコルサーバ・ライブラリの実装と、通常のファイル転送プロト コル(ftp)による輻輳時のジッタの評価結果について述べる。

Distributed multimedia systems must handle multiple continuous media streams at once. The network subsystem must thus be able to transmit aggregation of VBR streams effectively. Data smoothing and shaping is necessary for high quality transmission. Flexibility is also required to handle different types of data formats and contents. In this paper, we propose the VoR system, a network subsystem framework for VBR continuous media transmission. It controls stream multiplexing by the scheduler in the network driver along with the QoS coordinator middleware. The system provides a way for application programs to interact with the underlying network. The system is designed on the principle of policy and mechanism separation, which enables to flexibly adapt to application dependent inter- and intra-stream QoS tradeoffs. Our preliminary result using MPEG2 streams indicates that while simple multiplexing gains some bandwidth utilization, better multiplexing can be achieved by providing more elaborate data dependent policies.

In this paper, we propose a new and efficient fast retransmit scheme for wireless LANs. Wireless LANs are becoming popular, and providing and efficient protocol over a wireless link is important. In our experiments with 2.4 GHz WaveLAN, the poor TCP performance observed was attributed to frequent expiration of the retransmission timer. To avoid unnecessary expiration of the retransmission timer, we propose a scheme in which fast retransmission is performed efficiently. The proposed modification only affects a TCP sender, and our version of TCP with the change is interoperable with existing TCP implementations. The change is especially effective in wireless LAN environments, where we have demonstrated dramatic improvements in throughput, 10-15%, via experiments with 2.4 GHz WaveLAN.

In this paper, we describe the research topics and the goals of the Seamless Migratory Application Framework (SMAF) project. The project aims to provide the common framework for building future distributed embedded applications. In the future computing environment, various types of computers will be embedded in home and office appliances, which are connected through networks. Many kinds of application may run in variety of environment, and they need to aware and adapt to the computing environment and user preferences. The SMAF project focuses on how the applications need to execute efficiently in these environments, and we are building the SMAF system framework.

We propose a framework of Quality of Service (QoS) management inside a sending host that controls execution of sending threads in consideration of utilization of processor capacity and network bandwidth. To distinguish from flows which require only best-effort service, we call a flow which requires a specific rate of service ``reserved flow.'' To guarantee QoS of such reserved flow both in processor- and network-intensive cases in a sending host, processor capacity reserve is allocated such that the rate of each reserved flow is attained and non-conforming data are policed before they are transmitted. Processor Capacity Manager and the network device driver exchange information in a cooperative manner to support the rate adaptive allocation of processor capacity reserve.

Traffic management schemes such as Connection Admission Control (CAC), policing, and traffic shaping are important to provide multimedia communications with better Quality of Service (QoS). In the conventional model, admission control and policing are done at intermediate nodes, and traffic shaping is done at the edge of a network. However, QoS of communications should be defined between tasks or threads rather than between hosts. Therefore traffic management inside a host is as important as that in networks. We propose software-based traffic management architecture over a real-time microkernel. The architecture focuses on the interface between a network driver and user threads calling the driver. We categorized services of communication threads into three classes: Real-Time at Guaranteed Rate (RT-GR), Real-Time at Available Rate (RT-AR), and Best-Effort (BE). Our architecture is designed for an environment containing a mixture of these services. In the architecture, a sender periodic thread of RT-GR or RT-AR is executed such that the sending rate matches a user-specified rate. The network driver monitors the per-flow rate of injected data and discards the data if the injected rate exceeds the user-specified rate. To avoid the continuous discarding of data, the sending thread can adjust its sending rate by periodically looking at logged data concerning the rate. RT-AR service can achieve more than the specified rate when bandwidth is available. The scheme of software traffic management is effective in attaining higher throughput not only for full-duplex Ethernet but also for ATM because the difference of rate between hardware and software is reduced. In this paper, we describe the design and implementation of the software-based traffic management architecture on Real-Time Mach. The results of performance evaluations demonstrate that our traffic management scheme performs well for full-duplex Ethernet.

In this paper, we propose the Neighbor-state Based Queuing (NBQ) for improving the performance of flows sharing link bandwidth under various transfer protocols. Each router with NBQ informs the next upstream router of its link status. According to the received status, NBQ dynamically sets the packet discard ratio of the non-adaptive flows. NBQ reduces the wasted nonadaptive flows and provides available bandwidth for the adaptive flows. We have demonstrated dramatic improvements of adaptive flow's throughput using our scheme through simulations.