当前位置:首页 >> >>

scalable multimedia distribution system_图文

Computer Communications 24 (2001) 105±123


Using dynamic con?guration to manage a scalable multimedia distribution system q
F. Kon*, R.H. Campbell, K. Nahrstedt
Department of Computer Science, University of Illinois at Urbana-Champaign, 1304 West Spring?eld Avenue, Urbana, IL 61801-2987, USA 1 Received 15 March 2000; accepted 4 September 2000

Abstract Multimedia applications and interfaces will change radically the way computer systems will look like in the coming years. Radio and TV broadcasting will assume a digital format and their distribution networks will be integrated to the Internet. Existing hardware and software infrastructures, however, are unable to provide all the scalability, ?exibility, and quality of service (QoS) that these applications require. We present a framework for building scalable and ?exible multimedia distribution systems that greatly improves the possibilities for the provision of quality of service in large-scale networks. We show how to use architectural-awareness, mobile agents, and a CORBA-based framework to support dynamic (re)con?guration, ef?cient code distribution, and fault-tolerance. This approach can be applied not only for multimedia distribution, but also for any QoS-sensitive distributed application. q 2001 Elsevier Science B.V. All rights reserved.
Keywords: Multimedia distribution; Dynamic con?guration; Middleware; CORBA; QoS-aware resource management

1. Introduction Multimedia interfaces will play a fundamental role in human±computer interaction in next generation computer systems. Applications will gradually abandon dull interfaces based on text and static graphics and move towards a more interesting look based on audio, video, and animations. Within one decade, there will probably be no distinction among the telephone, radio, TV, and data networks; and in future decades, no distinction between computer and dedicated radio and TV receivers. Everything will be integrated on an extension of today's Internet. The Internet as we know today, however, is not suited for the distribution of high-quality multimedia to a large number of clients. A number of changes in the hardware and software that currently support the Internet will have to occur. We need systems to better control large numbers of multimedia ?ows, with support for Quality of Service (QoS) provision. In the past ten years, several research groups have worked
q This research is supported by the National Science Foundation, grants CCR 96-23867. 98-70736, 99-70139, and 99-72884EQ. Fabio Kon is supported in part by a grant from CAPES, Brazil, proc. 1405/95-2. * Corresponding author. E-mail addresses: f-kon@cs.uiuc.edu (F. Kon), roy@cs.uiuc.edu (R.H. Campbell), klara@cs.uiuc.edu (K. Nahrstedt). 1 http://choices.cs.uiuc.edu/2K.

on architectures for distributing digital multimedia content through the Internet and intranet infrastructures. In particular, researchers working on the technologies related to the MBone [1,2] and to adaptive audiovisual streaming protocols [3,4] provided signi?cant contributions. We noticed, however, that existing MBone and unicast solutions do not provide the degree of control, ?exibility, and the possibilities for QoS management that the next generation applications require. In this article, we present an architecture for scalable multimedia distribution that meets those requirements. The architecture combines modern technologies from different Computer Science ?elds and extends them when necessary. We applied the ideas presented in this article to build an object-oriented multimedia distribution system in C11 [5] and demonstrated that it is possible to use the existing Internet to distribute low and medium bandwidth multimedia to thousands of simultaneous users. Our early experiments, though, pointed out dif?culties in managing such a largescale system and keeping it available with an acceptable QoS. It showed the necessity for a better support for dynamic (re)con?guration, code distribution, and provision of fault-tolerance. This article describes our most recent achievement in this area, i.e. an integrated software architecture addressing the problems we encountered in the past. Our solution is based on an extensible Re?ector system, mobile agents,

0140-3664/01/$ - see front matter q 2001 Elsevier Science B.V. All rights reserved. PII: S 0140-366 4(00)00293-0


F. Kon et al. / Computer Communications 24 (2001) 105±123

. . .

Capture Station

Master Reflector


Public Reflector


Public Reflector

Reflector Public Reflector


. . .
Public Reflector

Capture Station

Master Reflector


Fig. 1. A Re?ector network distributing two video streams.

and a CORBA framework for dynamic con?guration and recon?guration. Section 2 presents an overview of the basic concepts involved in scalable multimedia distribution. Section 3 describes our early design, discussing its limitations. Section 4 describes our new, enhanced architecture with support for dynamic con?guration and explains the synergistic relationships between dynamic con?guration and QoS. Section 5 presents a concrete implementation of our architecture and Section 6 discusses performance evaluation. Finally, Section 7 discusses related work and Section 8 presents our conclusions and future work. 2. Basic concepts Before going into a more detailed description of our architecture, we ?rst present a brief overview of the most important concepts related to our approach to scalable multimedia distribution. (1) Re?ector. It is the key element of our distribution system. It acts as a relay, receiving input data packets from a list of trusted sources and forwarding these packets to other Re?ectors or to programs executed by end-users. The distribution system is composed of a network of Re?ectors that collaborate with each other to distribute the multimedia data over local-area, metropolitan-area, and widearea networks. (2) Re?ector administrator. A privileged user that is responsible for managing a Re?ector network. Different portions of the network may be managed by different administrators, making large systems more manageable and helping to deal with different security domains. (3) Dynamic (re)con?guration. As computer environments become more dynamic, a major requirement for next generation computer systems is the ability to customize

the system to the characteristics of the environment in which it executes. This property is known as dynamic con?guration. Also of extreme importance is the ability of a system to change, on-the-?y, its internal components and con?guration parameters to adapt to changes in the environment. This is called dynamic recon?guration 2. (4) Quality of service (QoS). Multimedia applications have stringent requirements with respect to computational resources such as memory, CPU, and network [6]. If these requirements are not met, the quality of the multimedia service is degraded. This quality can be measured according to different metrics. A videoconference service, for example, can be evaluated according to the size and the number of colors of the video frames, the audio sampling rate, the video frame rate and jitter, the synchronization of audio and video, the delay from the sender to the receiver, the number of frames that are lost or defective, and so on. (5) CORBA. The OMG Common Object Request Broker Architecture is an architecture for distributed object communication that is both language-independent and platform-independent. It is based on a standard interface de?nition language (IDL) and includes standard de?nitions for interoperable distributed communication [7]. It also de?nes standard interfaces for services such as naming, trading, real-time, persistence, and transactions [8].

3. Scalable distribution Using traditional, centralized video-on-demand servers, it is possible to stream video to hundreds of simultaneous clients [3]. Exploiting IP-Multicast [9], this number increases to several thousands or, maybe, a few millions.
2 For simplicity, the term dynamic con?guration may be used to denote both dynamic con?guration and dynamic recon?guration.

F. Kon et al. / Computer Communications 24 (2001) 105±123
IPMulticast/ local Ethernet


Reflector F TCP/ interstate fiber

TCP / longdistance fiber link Reflector A IPMulticast/ MBone


Master Reflector

Reflector B UDP/ ISDN

TCP/ phone dialup TCP/ transatlantic satellite link

Capture Station

Master Reflector TCP/ longdistance fiber link

Reflector C

TCP / transatlantic satellite link Reflector D Reflector E VDP/ phone dialup

IPMulticast/ MBone


Fig. 2. A heterogeneous Re?ector network.

However, when using IP-Multicast to send video to a large number of clients, one has little control over the transmission. It becomes dif?cult to provide support for a number of desired features such as QoS, security, accounting, and reliability. To address this problem, we developed a scalable multimedia distribution framework whose architecture is described in this section. 3.1. The Re?ector As noted above, the Re?ector is the key element of the distribution framework. It is a user-level program that, at the application level, performs activities similar to the ones performed by network routers at the hardware level (which is where IP-Multicast is implemented). Since it is implemented in software, not hardwired into the router, the Re?ector is more ?exible, easy to deploy and evolve, and can be dynamically customized to users, applications, and environments. Re?ector data packets are encoded with RTP, a userlevel protocol for real-time applications [10] de?ned by the Internet Engineering Task Force (IETF). RTP packets can be transmitted over different kinds of low-level transport protocols such as TCP, UDP, and IP-Multicast. The Re?ector combines data packets into logical groups called channels. It examines a 2-byte channel_ID ?eld in the packet header and forwards the packet to the clients and Re?ectors that subscribed to that particular channel. Each channel can contain multi-track data for different applications such as video and audio conferencing, live highbandwidth MPEG broadcasting with TV quality, stored low-bandwidth H.263 video and GSM audio, HTML

news, or current stock values. Clients and administration programs can connect to a Re?ector in order to get information about available channels, Re?ector load, bandwidth utilization, historical statistics, etc. The Re?ector network topology is determined by each Re?ector's con?guration. This information speci?es input and output connections, access privileges, maximum allowed number of users, etc. The information is stored in a database controlled by the Re?ector administrator. Fig. 1 depicts a generic Re?ector network that distributes two video streams in different channels. In this ?gure, two capture stations send their video streams to ?master? Re?ectors; the streams may traverse several ?intermediate? Re?ectors until they reach ?public? Re?ectors to which end-user clients can connect and receive the video streams. All the Re?ectors in the ?gure are initiated with exactly the same code. As explained in Section 4.1, the system then customizes each Re?ector according to their individual requirements. To provide fault-tolerance Re?ectors can accept redundant inputs for the same channel from several sources. It assumes that one of them is the primary source and ignores the data coming from other redundant sources. If an error in the connection with the primary source is detected or if it remains silent for a pre-de?ned period of time, the Re?ector automatically switches to the next source in its redundancy list. Software routing at the Re?ectors may introduce extra latency and jitter. Typically, latency will only be a problem for conferencing applications and only if a long chain of Re?ectors is used. Jitter can be minimized by reserving CPU and memory as described in Section 4.1.


F. Kon et al. / Computer Communications 24 (2001) 105±123

3.2. Data distribution protocols In order to support different types of inter-Re?ector data distribution protocols transparently, the Re?ector framework encapsulates the concept of a network connection into an abstract C11 class named Connection that de?nes the basic interface for all types of network connections used by the Re?ector. This abstract class implements some of its own methods, but the majority of the connection-related code is implemented by its subclasses: TCPConnection, UDPConnection, MulticastConnection, etc. Re?ector control information and meta-data is sent through reliable TCP connections. Multimedia data may be sent through unreliable connections in order to achieve a higher throughput. Also, different channels may use different connection types; for example, in a single Re?ector a video channel may use UDP while a text channel with news articles uses a TCP connection. Fig. 2 depicts a concrete example of a highly heterogeneous Re?ector network. In this example, the network distributes two audiovisual streams. The ?rst comes from a mobile camera, mounted on a helicopter, that sends its stream to a ?master? Re?ector over a wireless link. This kind of link presents a high rate of packet loss not related to congestion, which makes protocols like TCP perform poorly. Thus, it is desirable to use a protocol optimized for this kind of situation like, for example, WTCP [11]. The second stream is sent to its ?master? Re?ector through a dedicated ISDN line. To optimize the bandwidth, one can use UDP as the communication protocol since its overhead is lower than that of TCP and the link offers a low loss rate. Re?ector C sends its streams to Re?ector D over the public Internet through a transatlantic satellite link. Even though this is a high-bandwidth link, its loss rate may be high, thus it is more appropriate to use TCP. Re?ectors A and D introduce the video streams into two distant points of the global MBone. Selecting the appropriate protocol for each situation, the Re?ector administrators improve the quality of service offered to the end-users, optimizing the utilization of the available network bandwidth and minimizing packet losses. Most of the Re?ector's code deals with objects of type Connection and is not aware of the Connection's underlying implementation. The actual connection type is speci?ed when the connection is created and does not need to be known after that. This approach allows programmers to plug in customized Connection subclasses by providing their own implementation of the Open, Close, Bind, Connect, Send, and Receive methods. In this manner, it is possible to incorporate, into the Re?ector, Connection subclasses that implement different transport protocols (such as the VDP QoS-aware adaptive protocol for the Internet [3] and the xbind QoS-aware reservation protocol for ATM networks [12]). Developers also use this mechanism to implement Connection subclasses that perform various operations on the data such as encryption, transcoding,

mixing, and downsampling. Finally, one can create new Connection types by composing existing ones. For example, one can create a CryptoMulticast connection type ? that encrypts the data and sends it out using Multicast ? by composing a Crypto connection with a Multicast connection. 3.3. Experience and lessons learned This technology was utilized in the live broadcast of NASA's JPL Path?nder mission [13]. During this broadcast ? which lasted for several months ? more than one million live video sessions were delivered to dozens of different countries across the globe by a network of more than 30 Re?ectors spread across ?ve continents. The Re?ectors ran in ?ve different operating systems (Solaris, Linux, Irix, FreeBSD, and Windows) and transmitted their streams over different kinds of network links. End-users could watch the video stream simply by pointing their web browsers to selected locations, causing their browsers to download a Java applet containing the video client. The applet connected to the Re?ector automatically, received the video stream, decoded it, and displayed it to the user in real-time. During this broadcast, we experienced three major problems: 1. As the code had not been tested on such a large scale and on so many different platforms, we found many programming errors both in the Re?ector code and in the client applet. Surprisingly, ?xing the error was, sometimes, easier than updating the code in the dozens of machines that formed the distributed system. The same problem occurred when a new version of the Re?ector, with added functionality, was released. System administrators had to manually connect to dozens of machines, upload the new code, shutdown the old version, and start the new one. 2. Often, we had to recon?gure the re?ector network by dynamically changing the distribution topology, or by setting new values to the re?ector con?guration parameters (e.g. maximum number of users, number of multimedia channels). The con?guration information for the re?ectors was stored in a centralized location. After updating this centralized database, we had to connect to each of the re?ectors and instruct them to download their updated con?guration. This process was tiresome and error-prone. 3. The only mechanism the Re?ector provided to support fault-tolerance was the redundant inputs described in Section 3.1. But this mechanism leads to a large waste of bandwidth. The redundant streams are always transmitted even though they are seldom used. With this experience, we learned that a user-level Re?ector system is, indeed, a powerful tool for managing large-scale

F. Kon et al. / Computer Communications 24 (2001) 105±123


multimedia distribution. It gives Re?ector administrators a tight control over the distribution, allowing for a better control of QoS. It achieves that through the de?nition of the distribution topology, the selection of appropriate communication protocols, and the possibilities for limiting the number of clients according to the available resources. We also learned, however, that it was important to provide better mechanisms for distributed code updates, dynamic recon?guration, and fault-tolerance. In the next section we describe a novel architecture for multimedia distribution that addresses the problems encountered in the previous approaches, leading to a more ef?cient, manageable, and reliable system. 4. Architecture With the exponential growth in the number of workstations, laptop computers, and PDAs connected to the Internet, and with the extra levels of mobility and dynamism that these devices bring, it becomes increasingly important for multimedia systems to be able to cope with dynamic changes in the environment. They must react to variations in resource availability, dynamically adapting their algorithms, updating parts of the system, and replacing software components when needed. The Re?ector is no exception to this trend as it needs to cope with different kinds of devices that may connect to it. A client running on a PDA, for example, may need to dynamically add a proxy into the Re?ector so that it can transcode the multimedia stream into a format suitable for the PDA (as done in the PalmPlayer video system [14]). These new requirements, together with the lessons learned in the past experiences (see Section 3.3), point out the necessity of supporting dynamic con?guration of the Re?ector system. Although dynamic con?guration is a ?eld that has been studied intensively in the past decade [15], the dynamic con?guration of QoS-sensitive systems pose completely new challenges that have not yet been thoroughly explored. In this case, the system must not only be able to recon?gure correctly, but also be able to carry out this recon?guration with minimal impact on QoS. The synergistic relationships between dynamic con?guration and QoS are clear. Dynamic con?guration allows the

use of the best policies for each situation. For example, a mobile computer displaying a video clip to its user could use a protocol optimized for wireless connections when the computer is using a wireless link, but dynamically recon?gure itself to use a TCP connection when the computer is hooked to a wired Ethernet connection. However, if the recon?guration process, itself, affects the quality of service negatively, it may not be worthwhile to do any recon?guration at all. Coming back to the example, if the dynamic recon?guration to the TCP connection is so expensive that the video is interrupted for several seconds, it is better to keep using the wireless link even when the wired link becomes available. We, now, present an enhanced architecture that greatly improves the possibilities for QoS management with new mechanisms for automatic con?guration (Section 4.1), scalable code distribution and dynamic Re?ector recon?guration (Section 4.2), and dynamic recon?guration of the network topology to provide fault-tolerance (Section 4.3). Furthermore, the dynamic con?guration is carried out with no negative impact on the QoS of the multimedia streams. 4.1. Automatic con?guration To solve the problem of maintaining the Re?ector instances up-to-date as the code of the Re?ector program evolves and to customize each Re?ector according to its role, we adopted the automatic con?guration approach [16,17]. The Re?ector is divided into a minimal core and components that are linked to the core at runtime to extend its functionality. These components may include the implementation of the different data distribution protocols and mechanisms for encryption, access control, accounting, down-sizing, and transcoding. The core is the only code that is initially shipped to the nodes that are going to execute the Re?ector. This is enough to initiate the bootstrapping process. The con?guration information for all the Re?ectors in a certain administrative domain are stored in a con?guration service. Fig. 3 depicts a schematic view of the automatic con?guration process, which is driven by a module of the Re?ector core called dynamic con?gurator. All the code for con?guration is encapsulated in the dynamic con?gurator. When a Re?ector starts, it ?rst contacts the con?guration

1 inputs Reflector core 4 3 QoSAware Resource Management Dynamic Configurator 5 2 Configuration Service

Fig. 3. Automatic con?guration architecture.


F. Kon et al. / Computer Communications 24 (2001) 105±123

4.2. Scalable code distribution and dynamic recon?guration The automatic con?guration process described above consists of a pull-based approach for code updates and con?guration. In other words, the Re?ectors take the initiative to pull the code and con?guration information from a central location. In large-scale systems, this may not be enough. We also need a mechanism to allow system administrators to push code and con?guration information into the Re?ector network ef?ciently. Our architecture achieves that by using the concept of mobile agents [21]. Using a special con?guration agent builder tool, Re?ector administrators are able to create a mobile agent and inject it into the Re?ector network. The agent traverses the Re?ector network and, in each node, is able to perform three kinds of actions: recon?gure the Re?ector changing its working parameters (e.g. maximum number of clients, security keys, access control lists), install the code for new components (e.g. new data distribution protocols, security mechanisms), and collect information about the Re?ector dynamic state (e.g. bandwidth utilization, current number of clients, total number of clients since startup). After visiting the nodes speci?ed by the administrator, the agent returns and displays the results of its actions and the collected information. Although the introduction of mobile agents promotes signi?cant bene?ts, it also brings additional security concerns. This is especially worrisome for QoS-sensitive applications since malicious or defective agents may utilize excessive resources and degrade the Re?ector QoS. Luckily, the problem of mobile agents security has been studied extensively in recent years and good results are starting to appear. In the Re?ector case, it is important to (1) restrict the sources of mobile code to trusted, authenticated entities [24], (2) limit the mobile code's resource consumption [22], and (3) limit its capabilities by con?ning it to secured environments like sand boxes for Java byte code or the Janus sand box for native code [23]. 4.3. Fault-tolerance Reliability is an important aspect of QoS. It is dif?cult to maintain a system working properly in the presence of network failures, node failures, software failures, and partial system shutdowns for maintenance. In addition, keeping the desired level of QoS irrespective of all these disruptive events is a major challenge for modern QoS-sensitive systems. The architecture addresses this problem by using architecturally aware dynamic recon?guration based on a framework for representing dependencies in distributed systems [16,17]. In multimedia distribution, the goal with respect to faulttolerance is to maximize availability without relying on redundant transmissions, which lead to waste of bandwidth. To achieve that, the architecture supports the dynamic recon?guration of the Re?ector inter-connections when

Fig. 4. Low-level QoS speci?cation using the Simple Prerequisite Description Format.

service to retrieve its con?guration (step 1 in Fig. 3). The con?guration service provides three kinds of con?guration information to the Re?ector: 1. A list of the components that must be dynamically linked to the core to customize this instance of the Re?ector. 2. A speci?cation of the QoS parameters required by this Re?ector. 3. The input and output connections to be created at startup. In step 2, the Re?ector not only receives its con?guration information, but also the actual code for the components to be linked to the core. In step 3, it passes the QoS speci?cation to an underlying QoS-aware resource management service that is responsible for managing local machine resources such as CPU and memory, and, when possible, distributed resources such as the network. The QoS speci?cation may be as simple as an SPDF speci?cation [17] like the one in Fig. 4, or as elaborated as a QML speci?cation [18]. The QoS-aware resource management service can be provided by systems like SMART [19] or the Dynamic Soft Real-Time Scheduler (DSRT) [20]. DSRT, for example, can use the QoS speci?cation to perform QoS-aware admission control, negotiation, reservation, and scheduling. Since the architecture relies on user-level processing of data packets, the use of a soft real-time scheduler is important to guarantee that the Re?ectors do not impose additional jitter to the multimedia streams. The con?guration is completed in step 4, when the dynamic con?gurator instructs the Re?ector to open the required input and output connections. Afterwards, when the Re?ector is running, the resource management service may detect exceptional changes in the resource availability or violations of the QoS requirements that may require modi?cations in the Re?ector con?guration. In such cases, it issues a call-back to the dynamic con?gurator with information about the exceptional conditions (arrow labeled 5 in Fig. 3). This automatic con?guration process greatly simpli?es system management since the Re?ectors always download the most up-to-date version of each component from the con?guration service. In that way, it eliminates the need to upload components to the entire network each time a component is updated.

F. Kon et al. / Computer Communications 24 (2001) 105±123


Name Service

Configuration Service

Component Repository

2 1
Source Reflector





9 8 DSRT

Reflector 5
Component Cache


Client Reflector

Fig. 5. Bootstrapping a Re?ector.

failures occur. The distributed system has knowledge of its own structure and is able to establish alternative routes to deliver the streams to its users. Furthermore, whenever possible, the system performs these recon?gurations without affecting the QoS perceived by its users. 4.3.1. Fault-recovery models In order to build an alternate distribution topology, the system must store enough information so that alternate routes can be found when failures occur. The question is: where to maintain this information? We considered, initially, a solution in which all the information about fault recovery would be placed in a centralized database accessible by every Re?ector. When a Re?ector R1 detects that one of its inputs failed or that it is silent for too long, it would contact a con?guration server with access to the centralized database and request the address of a re?ector R2 from which it could receive the missing streams. The con?guration server would return this information and contact R2, recon?guring it to send data to R1. The advantage of this approach is that very little information is stored on the Re?ectors, all the fault-recovery information and policies are centralized in a single location, facilitating the manipulation of this information by a single entity: the con?guration server. The second solution is to store fault-recovery information in the Re?ectors. That is, each Re?ector would store a list of alternate Re?ectors that could be used as backups. The advantage of this approach is that it does not lead to a single point of failure and does not impose an extra load on a possibly already overloaded con?guration server. This solution may be more dif?cult to implement but it tends to be more scalable. We believe that the optimal solution to this problem is one that encompasses both models. On the one hand, each Re?ector should have some knowledge about its local surroundings and should be able to perform recon?gurations

by communicating with its immediate neighbors, without being a burden to the centralized con?guration server. On the other hand, the con?guration server should maintain global knowledge about the system topology. This centralized, global knowledge should be used not only as backup, in case the Re?ector's localized information is not enough to keep the system functioning, but also to perform optimizations such as dynamic changes in the network topology to improve the quality of service and promote load balancing. Our architecture adopts the hybrid model described above. It distributes the knowledge throughout the Re?ector network and makes each Re?ector aware of its dependence relationships with other Re?ectors. Thus, the Re?ectors are able to make recon?guration decisions by themselves, without relying on a centralized entity. In addition to this, the global system topology is maintained in the con?guration service so that a Re?ector administrator or an ?intelligent? software module can perform global optimizations in the distribution network. Each Re?ector contains an object of the type ComponentCon?gurator [16] which stores a list of entities on which the Re?ector depends (its source Re?ectors) as well as a list of entities that depends upon the Re?ector (client Re?ectors or end-user clients). In addition, a ComponentCon?gurator may also contain a list of alternatives, which could serve as alternate inputs for the multimedia streams in case the input for a given channel becomes silent. 5. Implementation We implemented a prototype of the architecture described in the previous section as a CORBA-based framework, which brings two immediate bene?ts. First, we are able to utilize a number of standard CORBA services implemented by the CORBA community such as the Naming Service and interact with existing CORBA tools. Second, we have implemented our mechanisms for automatic con?guration and dynamic recon?guration as CORBA


F. Kon et al. / Computer Communications 24 (2001) 105±123

Fig. 6. (a) Screen shot of the Name Service GUI. (b) Agents GUI.

services exporting well-de?ned IDL interfaces. This makes them easily accessible to other systems and reusable in other contexts. 5.1. Implementing automatic con?guration The ?rst major change we had to do in the Re?ector implementation to accommodate the new design was the adoption of a component-oriented model. We reorganized the implementation of the Re?ector program, breaking it into dynamically loadable components. Surprisingly, this proved to be not so dif?cult, thanks to the original objectoriented design of the Re?ector that was based on loosely coupled objects interacting via well-de?ned interfaces. This component-based model facilitates the customization of the Re?ector program at startup, allowing the system to select the components that are best suited for a speci?c environment at a certain time. It also facilitates the dynamic recon?guration of running Re?ectors to adapt to changes in the environment and to install new versions of components onthe-?y. Fig. 5 presents a schematic overview of the Re?ector bootstrapping process in which the Re?ector con?gures itself. At startup time, each Re?ector contacts the CORBA Name Service to locate the con?guration service (steps 1

and 2 in Fig. 5). From the con?guration service, it retrieves its speci?c con?guration information, which may contain three kinds of speci?cations: (1) the components that must be dynamically loaded to build this instance of the Re?ector; (2) the amount of physical memory and the share of the CPU that should be reserved for this instance of the Re?ector; and (3) the input and output connections to be created at startup. The Re?ector uses the information of the ?rst kind (which is obtained in step 3) to contact our CORBA Component Repository and fetch the code implementing the desired components 3 (step 4). It then dynamically links these components into the runtime system (step 5). If con?guration speci?cations of the second kind are present, they are fetched in steps 6 and 7. Next, the Re?ector contacts the underlying middleware to request the reservation of the required resources, which is achieved with the help of the Dynamic Soft Real-Time Scheduler (DSRT) [20]. After loading all the required components, the Re?ector registers itself with the Name Service, so that it can be easily located by other system entities, and opens all the input and

3 To minimize startup time and network load, the components fetched from the Component Repository can be cached locally.

F. Kon et al. / Computer Communications 24 (2001) 105±123


output connections using the speci?ed protocols (steps 9 and 10). Administrators of the Re?ector system can look at the available broadcast and videoconference sessions by using a graphical user interface that interacts with the CORBA Name Service. Fig. 6(a) is a screen shot that exempli?es the use of this interface. It shows three independent Re?ector networks: the ?rst called VirtualMeetingRoom reserved for audio and videoconferencing, possibly divided according to interest groups; the second called News that could contain several news channels; and the third called OlympicGames that could contain audio and video broadcast channels related to Olympic events. The administrator can also see that there are three available component repositories for three different kinds of architectures and can use the GUI to upload, download, and remove Re?ector components from the repositories. Finally, the administrator can see that the OlympicGames network is composed of the ?ve re?ectors shown in the upper right-hand window. The CORBA IOR (Interoperable Object Reference) of the Re?ector at delirius.cs.uiuc.edu is shown in the bottom right-hand window. 5.2. Mobile recon?guration agents To support ef?cient code distribution and the dynamic recon?guration of large-scale Re?ector networks, we use a generic middleware infrastructure for mobile agents [24]. This infrastructure is based on dynamic TAO [25], an enhanced CORBA Object Request Broker (ORB) that exports a recon?guration interface that can be used to (1) upload executable code to remote ORBs, (2) dynamically load components into the system runtime, (3) change the con?guration parameters of application components, and (4) modify the internal architecture of component-based applications. The infrastructure includes the support for recon?guration agents which work as ?smart capsules'' that traverse a network of distributed ORBs speci?ed by the application administrator. These agents contain a directed graph representing the order in which the ORBs are to be visited and a script (or Java program) specifying recon?guration or inspection commands that are processed by each of the ORBs in their path. The results are collected and returned to the application administrator by following the reverse direction in the directed graph. This infrastructure was all we needed to solve the two initial problems (see Section 3.3) in the previous Re?ector version. The recon?guration interface supports the operations upload_implementation and con?gure_implementation, which can be used for code distribution and for sending con?guration messages to application components, respectively. Using a graphical user interface, the Re?ector administrator can build a recon?guration agent that carries new implementations of Re?ector components. One of the GUI

windows (see Fig. 6(b)) allows the administrator to draw the vertices and edges of a directed graph that speci?es the collection of Re?ectors that will receive the new code and the path through which the code will be transmitted. Each node in Fig. 6(b) refers to a Re?ector running on a different location. The directed edges specify the paths the agents traverse. Using the GUI, the Re?ector administrator can send state inspection or recon?guration commands to subsets of the Re?ector network. The operation results are combined and returned to the administrator who can verify the results of all operations in the Re?ectors and take care of individual failures, if any. 5.3. Supporting fault-tolerance The architecture supports fault-tolerance by using ComponentCon?gurator objects to represent the dependencies between Re?ectors. When failures occur, the system uses the dependence information to locate alternate routes and keep the system functioning. The ComponentCon?gurator implementation stores the dependencies as a list of CORBA IORs, which allows for prompt communication no matter where the objects are located. A subclass of ComponentCon?gurator, called Re?ectorCon?gurator, contains the policies for reshaping the network topology in case of failures and encapsulates all the code to deal with these recon?gurations. This approach proved to be very effective in keeping a clear separation of concerns in the Re?ector code. The classes that deal with the Re?ector's normal operation are totally unaware of the Re?ectorCon?gurator and of any code that deals with recon?guration. This clean separation also makes it easy to plug different kinds of Re?ectorCon?gurator to support different recon?guration policies. 5.3.1. Triggering recon?gurations Four kinds of events can trigger dynamic recon?guration: 1. A Re?ector shutdown message sent by the Re?ector administrator or a kill command executed by the local system administrator. 2. Software errors that lead to a segmentation fault or bus error. 3. A recon?guration order sent by the Re?ector administrator. 4. Sudden machine crashes or network disconnections. In the ?rst two cases, the Re?ector captures those events using signal handlers installed with the UNIX signal function or the Windows SetConsoleCtrlHandler function. In the UNIX implementation, for example, the administrator can kill a Re?ector by pressing Ctrl-c on the terminal executing the Re?ector, by sending a shutdown message to the re?ector using telnet, or by using the kill command. The Re?ector captures the events generated by Ctrl-c, kill,


F. Kon et al. / Computer Communications 24 (2001) 105±123


Reflector A Reflector C

Reflector D


Server (a) Server

Reflector B

Reflector E


Reflector A

Reflector D


Server (b)

Reflector B

Reflector E


Fig. 7. (a) Network with 5 Re?ectors. (b) After Re?ector C is killed.

segmentation faults, and bus errors by implementing signal handlers for the SIGINT, SIGTERM, SIGSEGV, and SIGBUS signals, respectively. In the third case, the Re?ector contacts the con?guration service to retrieve its new con?guration information and reprocesses it, recon?guring its input and output connections. Finally, the fourth case is the only one in which it is not possible to keep the client multimedia stream uninterrupted without relying on redundant streams to the same Re?ector. The solution in this case is to detect when the input for a given channel has failed or has been silent for too long and then locate an alternative input either by using the local list of alternatives or by contacting the con?guration server. As described below, our current implementation focuses on supporting dynamic recon?guration in the presence of the ?rst two kinds of events. 5.3.2. The recon?guration process When an administrator (or other system entity) requests to kill a Re?ector, the system executes a special event handler called abandonRe?ectorNetwork. This handler takes the following three actions. 1. Unregisters the Re?ector from the Name Service. 2. Using CORBA, sends a FINISHED event to the Re?ectorCon?gurators of all the sources (inputs) of this Re?ector; the event carries a list of the clients of the ?nishing Re?ector. 3. Sends a FINISHED event to the Re?ectorCon?gurators of all the clients (outputs) of this Re?ector, carrying a list of its sources. When a Re?ectorCon?gurator receives a FINISHED event from a source Re?ector, it adds all the Re?ectors in the list of sources of the ?nishing Re?ector to its list of inputs. Conversely, when a Re?ectorCon?gurator receives a FINISHED event from a client Re?ector, it adds all the

Re?ectors in the list of clients of the ?nishing Re?ector to its output list. Fig. 7(a) shows a sample Re?ector network where Re?ector C has two inputs and two outputs. When C is killed and the recon?guration process described above completes, the new con?guration becomes the one in Fig. 7(b). In order to be able to carry out the recon?guration without any glitches in the multimedia streams and without affecting the system quality of service, we had to adopt a multi-threaded solution which we describe in Section 6.3. 5.3.3. The recovery process When a Re?ector starts its execution for the ?rst time or when it is restarted after being shutdown for some reason, it executes an initialization process. In this process, in addition to performing the actions described in Section 5.1, it performs the following three actions: 1. Registers the Re?ector with the Name Service. 2. Using CORBA, sends a STARTED event to the Re?ectorCon?gurators of all the clients (outputs) of this Re?ector; the event carries a list of the sources of the new Re?ector. 3. Sends a STARTED event to the Re?ectorCon?gurators of all the sources (inputs) of this Re?ector, carrying a list of its clients. Upon receiving a STARTED event from a new source Re?ector, the client Re?ector opens a new input connection to the new Re?ector. If it is also receiving input from one of the sources of the new Re?ector, it closes that input connection as soon as the data from the new source is available. An analogous process happens upon receiving a STARTED event from a new client Re?ector. These mechanisms allow the distribution system to recover its original topology after a faulty re?ector restarts. Therefore, if the system con?guration is the one in Fig. 7(b) and the Re?ector C

F. Kon et al. / Computer Communications 24 (2001) 105±123


Fig. 8. The TV client Applet.

recovers, then the con?guration automatically switches back to the one in Fig. 7(a). Note that we do not have an automatic mechanism for restarting faulty Re?ectors. This requires the administrator's intervention, which seems to be natural since a Re?ector goes out of service either because of an administrator's command or because of a failure in the system. In both cases, the administrator's attention is advisable. Alternatively, if desired, the Re?ector can be added to the list of daemons that are executed when a machine boots, eliminating the need for manual intervention when a machine crashes and restarts. 6. Performance evaluation In this section, we describe the experiments we performed with the Re?ector system in the course of the last four years and draw conclusions about its performance. The experiments are divided into the following groups. (1) Evaluating the system's reliability and usability in large-scale, wide-area broadcasts. (2) Measuring the capacity of a single Re?ector with respect to the maximum number of clients and maximum bandwidth it can support. (3) Evaluating the impact of recon?guration of the network topology on the QoS perceived by the end-users. (4) Measuring the performance improvements obtained by using mobile agents for code distribution and recon?guration of Re?ector networks. 6.1. Large-scale, wide-area broadcast In March 1997, we performed the ?rst experiments with the initial implementation of the Re?ector. At that time, we

ran our TVStation application in three SPARCstations 20 running Solaris 5.5 and using the SunVideo interface for video capturing. Using an H.263 software encoder, the TVStation was able to generate an average of three frames per second while still allowing the machines to be used as personal workstations. For more than one month without interruption, the machines sent live video from our of?ces to a small Re?ector network. Users could receive the video by pointing their Web browsers to our personal home pages on the web. An applet implementing an H.263 decoder was automatically loaded by the web browser and was able to receive the video even over standard telephone lines. Since H.263 decoding is much less computationally intensive than encoding, a simple 133 MHz PC was able to decode and display the three video sessions simultaneously, totaling more than nine frames per second. Although rather small in the number of Re?ectors and bandwidth, this initial experiment showed that it was possible to keep a Re?ector network running for several weeks without any interruption. Our technology was later chosen to broadcast, live over the Internet, the NASA JPL Mars Path?nder mission [13,5] 24 h a day from July to October 1997. The capture station at NASA's Jet Propulsion Laboratory was composed of a 133 MHz Pentium-based PC equipped with an Osprey1000 card for capture and encoding H.263 video. The data rate was set to 24 kbps so the stream could be received by 28.8 kbps modems. It contained half-rate GSM audio and 3±5 frames per second of H.263 video, depending on the encoded images. Fig. 8 shows the TV client applet running within Netscape. In order to carry out a broadcast of such an enormous magnitude and avoid the complete collapse of our networks, we requested collaboration from other institutions. In a few days, we were able to establish a global network utilizing resources from several universities, research laboratories, space agencies, Internet service providers, computer manufacturers, and mass media corporations. Our group organized a distribution tree composed of more than 30 Re?ectors spread across ?ve continents. Fig. 1 depicts a network similar to the one we utilized. ?Master'' re?ectors received data from the capture station and forwarded them to ?intermediate'' Re?ectors across the globe. These Re?ectors forwarded the data to ?public'' Re?ectors which were associated with Web pages and accepted connections from end-user applets. This scheme provided live audio and video to dozens of countries around the globe. Each of the Re?ector sites was capable of serving from 30 to 300 simultaneous clients depending upon the bandwidth of its connection to the Internet. This network could potentially serve more than 3000 simultaneous clients 24 h a day. During the initial four days of the broadcast, the Re?ector network was able to deliver half a million multimedia streams. Within two weeks, usage climbed to more than one million video and audio sessions. After the ?rst few weeks, interest in the Path?nder mission decreased, but


F. Kon et al. / Computer Communications 24 (2001) 105±123


Reflector A

Reflector B

Reflector C


Fig. 9. Recon?guration experiment testbed.

the broadcast of the NASA Select TV over the Internet continued for several months. During the course of the broadcast, some recon?guration of the network topology was required in order to accommodate changes in the availability of Re?ector hosts. We had to redirect streams to different primary Re?ectors and often add or remove public Re?ectors. All these modi?cations were performed remotely without stopping the transmission. 6.2. Single re?ector capacity The only bottleneck in the NASA experiment was the bandwidth of Internet connections on the sites hosting Re?ectors. In order to guarantee a good quality of service to our clients, we were forced to deny approximately one million connection requests. During a different broadcast (featuring the Indy500 race), a single Re?ector running on a Sun Ultra 2 was capable of serving nearly 800 simultaneous clients. This showed that a much larger number of users could be serviced if enough bandwidth were available. Recent experiments in our laboratory show that a Re?ector running on a Sun Ultra-60 with Solaris 7, two 360 MHz processors and 1280 MBytes of RAM requires less than 30% of one of the processors' time to stream a 26 kbps video to 1020 simultaneous clients using TCP over a 100 Mbps Fast Ethernet. The bottleneck in this experiment is the maximum number of sockets that a Solaris process is allowed to open in our test machine: 1024. In machines that execute Solaris 7 in 64-bit mode, this limitation no longer exists. The same machine transmitting a 1 Mbps stream to 85 clients, uses less than 10% of one of its processors' time. The bottleneck in this case is the network which is not able to sustain a TCP payload data rate of more than 85 Mbps. Assuming that enough machines with high-bandwidth connections were available, we could use the software technology described in this article to build, for example, a network of 300 Re?ectors serving 1000 clients each, totaling 300,000 simultaneous unicast clients. Since the Re?ector also supports IP-multicast, the actual number of clients could be much larger. With large networks composed of more than 100 Re?ectors, the problem becomes the manageability of such a large

number of Re?ectors from a single point. We developed the mechanisms for automatic recon?guration of the Re?ector network in case of failures to address this exact problem. Even with this automated facility, it is advisable to divide the Re?ector network in collaborating groups of 100±200 Re?ectors, each group being managed by a different administrator. 6.3. Recon?guration impact on the QoS More recently, we conducted experiments to evaluate the impact of the automatic recon?guration of the Re?ector network on the quality of the multimedia streams received by the end-users. We carried out this set of experiments on seven Sun Ultra machines (two Ultra-60 and ?ve Ultra-5) connected by a 100 Mbps Fast Ethernet. As depicted in Fig. 9, three of these machines executed our Re?ector program. The fourth machine executed TestServer, a program we created to synthesize an RTP stream with bandwidth and packet rate speci?ed in its command line. A ?fth machine executed TestClient, a program that receives the packets from a Re?ector and logs the packet arrival times into a ?le. The sixth machine executed the CORBA Name Service and the last one, the con?guration service and component repository. As explained in Section 5.3, when the Re?ector B goes down, the system recon?gures itself automatically and adopts the new topology shown in Fig. 10. When the Re?ector B recovers, the system recon?gures itself back to the topology shown in Fig. 9. We evaluated the impact of these recon?gurations on the QoS perceived by the endusers by using the information in the log ?le to compute the packet inter-arrival time at the TestClient. The following ?gures plot packet inter-arrival time (in seconds) over time (in seconds). Each experiment lasted for 16±18 s. Fig. 11 shows the packet inter-arrival times at our testbed client when no recon?gurations take place and with the TestServer sending an RTP/UDP stream of 1.2 Mbps at 30 packets per second. The TestServer program uses an adaptive algorithm to keep its average output bandwidth as close as possible to the output bandwidth speci?ed at its command line. Once every K seconds, the program checks its output


Reflector A

Reflector B

Reflector C


Fig. 10. Recon?guration when Re?ector B goes down.

F. Kon et al. / Computer Communications 24 (2001) 105±123


Fig. 11. No recon?gurations. Bandwidth ? 1.2 Mbps, packet rate ? 30 pps.

bandwidth and modi?es its packet inter-transmission time to adapt to the changes in its measured output bandwidth. The packet size is ?xed throughout each experiment and it is computed by dividing the bandwidth by the packet rate. In our ?rst experiments, we set K ? 1 and the desired packet rate to 30 packets per second. The TestServer, then, tries to send one packet every 33.3 ms. But, since the Solaris clock tick is set to 10 ms, it does not let a program sleep for exactly 33.3 ms. Hence, our adaptive program ends up sending one packet every 30 or 40 ms which explains most of the variations throughout the experiment in Fig. 11. The small and the localized variations (such as the one at 8.5 s) are due to changes in the network and machine loads caused by other applications. Fig. 12 shows the impact of recon?gurations in our ?rst implementation of the fault-tolerance mechanism. The continuous arrows point to the instants in which the Re?ector B is killed and the Re?ector network topology switches from the con?guration in Fig. 9 to the one in Fig. 10. The dashed arrows point to the instants in which the Re?ector B recovers and the topology switches back to the one in Fig. 9. One can see that when we killed the Re?ector, there was a big peak in the inter-arrival time at the client. This happened because, in that implementation, as soon as the Re?ector received the termination signal, it stopped forwarding packets and sent events to its neighbors announcing that it was going to shutdown. This caused a delay until the other Re?ectors were able to recon?gure their inputs and outputs. In a wide-area network, the delays would be even larger and the QoS perceived by the end-user would be degraded. We solved this problem by creating a new thread to manage the recon?guration. While the new thread contacts

the Re?ector's sources and destinations to announce the end of the service, the old thread continues to perform the Re?ector's normal packet-forwarding operations. The Re?ector only shuts itself down after it receives a con?rmation from its neighbors that its service is no longer needed. If this con?rmation does not arrive, the Re?ector administrator still has the chance to kill the Re?ector anyway at his or her own discretion. Alternatively, the Re?ectorCon?gurator can be programmed to timeout after a given period so that one does not need to rely on administrator intervention. One can observe another problem in the graph of Fig. 12: as the Re?ector recovers, there is a sudden change in the packet inter-arrival time, which approaches zero for a couple of packets in a row. This happens because the enduser receives some repeated packets that are sent both by the ?backup'' connection being deactivated and the new, ?recovered'' connection. This problem of duplicated packets is solved very easily by implementing a new subclass of Connection that uses the RTP sequence number to drop repeated packets. Fig. 13 shows the result of running the same experiment after implementing these two improvements. One can see that the recon?gurations take place without affecting the QoS perceived by the end-user (there is no correlation between the arrows in Fig. 13 and the variations in packet inter-arrival times). Furthermore, since the old Re?ector keeps a thread performing the Re?ector's normal operations until the recon?guration is completed, the quality of service is not degraded even in wide-area networks where the latency to contact the neighboring Re?ectors may be large. The lack of any correlation between the Re?ector recon?gurations and the jitter (i.e. the variance in the packet


F. Kon et al. / Computer Communications 24 (2001) 105±123

Fig. 12. Re?ector recon?gurations. Bandwidth ? 1.2 Mbps, packet rate ? 30 pps.

Fig. 13. Re?ector recon?gurations. Bandwidth ? 1.2 Mbps, packet rate ? 30 pps.

F. Kon et al. / Computer Communications 24 (2001) 105±123


Fig. 14. High network and CPU load. Bandwidth ? 1.2 Mbps, packet rate ? 30 pps.

inter-arrival time) is even more clear in Fig. 14 that shows a similar experiment carried out in a period of high network and machine loads. In this case, we set K to 10 so that the TestServer would keep its transmission rate constant for longer periods. Finally, Fig. 15 shows one more experiment in which we

used the Re?ector network to distribute an RTP stream generated by the vat audioconference tool for the MBone. The acoustic perception of the users listening to audioconference con?rms what one sees in the graph: the recon?gurations do not degrade the quality of service of the audioconference.

Fig. 15. Re?ector recon?gurations. vat bandwidth ? 45 kbps, packet rate ? 25 pps.


F. Kon et al. / Computer Communications 24 (2001) 105±123

40 35

distribution tree point to point

RoundTrip Time (s)

30 25 20 15 10 5 0 0.5 1 2 4 8 16 32 64 128 256

(a) 1.7 1.6 1.5

Component Size (KBytes)
distribution tree point to point

RoundTrip Time (s)

1.4 1.3 1.2 1.1 1 0.9 0.8 0.7 0.6 1 2 3 4 5 6 7 8


Number of Commands in Agent
Fig. 16. (a) Uploading component to 9 nodes. (b) Agent visiting 9 nodes.

6.4. Code distribution and dynamic recon?guration with mobile agents In order to evaluate the response time and relative performance gains made possible by our mechanisms for Re?ector code distribution and recon?guration based on mobile agents, we established an intercontinental testbed with the collaboration of the Departments of Computer Science at the Rey Juan Carlos University in Spain and at the Campinas State University in Brazil. The testbed consisted of the following three groups of machines. (1) Two Sun Ultra-60 and one Sun Ultra-5 machines running Solaris 7 in the cs.uiuc.edu domain, (2) three 333 MHz PCs running Linux RedHat 6.1 at escet.urjc.es, and (3) three 300 MHz PCs running Linux RedHat 6.1 at ic.unicamp.br. The machines inside each group were connected by 100 Mbps Fast Ethernet networks while the groups were connected among themselves through the public Internet. We executed nine instances of our middleware, one in each node, and injected different kinds of agents in this

network. To avoid drastic oscillations in the available Internet bandwidth and latency, and to minimize undesired interference, we carried out the experiments during the night 4. We measured the average bandwidth between our laboratory in Illinois and the remote ones by transferring a 100 Kbytes ?le via FTP ?ve times and measured the latency by using the ping command. The average bandwidth and round-trip latency between our lab at cs.uiuc.edu and the nodes at escet.urjc.es were 76 KBps and 170 ms, respectively. Between our lab and ic.unicamp.br, 32 KBps and 270 ms, respectively. To measure how the bene?ts of using a distribution tree similar to the one in Fig. 6(b) varies with the size of the agent, we sent a series of agents carrying the code for components to be installed in the remote nodes. As shown in Fig. 16(a), as the size of the component being uploaded
4 When we ran the experiments during times of high network traf?c and congestion, the performance numbers were even more in favor of our mobile agents approach.

F. Kon et al. / Computer Communications 24 (2001) 105±123


increases, the relative gain of using a distribution tree instead of point-to-point connections to each node increases signi?cantly. Fig. 16(b) shows the elapsed time for executing agents carrying from one to eight recon?guration commands. Each point in the ?gures represents the arithmetic mean of 10 runs of each experiment. The vertical bars represent the standard deviation. These results demonstrate clearly that our approach for code distribution and recon?guration based on mobile agents can provide extreme performance improvements and better predictability in the management of wide-area distributed QoS-sensitive systems. We expect it to be of great help in the management of our multimedia distribution system. 6.5. Summary The experiments described in this section let us draw the following conclusions about the Re?ector system. 1. It provides a good level of scalability for multimedia distribution over wide-area networks and is able to keep the service running for several months without interruption. 2. It supports, at least, 1000 unicast clients per Re?ector depending upon the available network bandwidth. In our Ethernet testbed, each Re?ector is able to handle streams with a TCP payload data rate up to 85 Mbps. 3. Its mechanism for dynamic recon?guration of the distribution network to deal with Re?ector failures is ef?cient and does not impact the quality of service as perceived by its end-users. 4. Its mechanisms for scalable code distribution and dynamic recon?guration based on mobile agents provide signi?cant performance improvements when compared to traditional client/server mechanisms.

7. Related work Our research bene?ts from and builds on previous work on IP-Multicast [9] and the Internet MBone [1]. The MBone is a virtual network that is layered on top of the Internet to support routing of IP-Multicast packets. It is composed of islands that can directly support IP-Multicast linked by virtual point-to-point links called tunnels. The system described in this article can be seen either as extending the MBone capabilities or (since we also support IPMulticast), as a set of tools for managing MBone broadcasts. The MBone relies on a multicast distribution engine that is hard-wired into the networking infrastructure, making it dif?cult to deploy new technologies for distribution. Our approach not only uses user-level entities that can be replaced easily, but it also supports dynamic recon?guration of running systems, allowing for easy incremental

evolution of the communication mechanisms with respect to QoS, security, and reliability. Amir, McCanne, and Katz developed an active service framework [26] that provides support for the dynamic instantiation of server agents, providing the desired service, in a collection of distributed nodes running their host manager (HM) daemon. They used this framework to implement a Media Gateway (MeGa) service [2]. As our Re?ector, the MeGa service can be used to transcode multimedia streams. Indeed, their research has focused on algorithms for ef?cient transcoding and down-sizing and, more recently, on adaptive bandwidth allocation algorithms and on the dynamic instantiation of media gateways between MBone sessions to deal with heterogeneous networks. Our research on the Re?ector architecture shares some concerns with their active service framework (which has some similarities with our automatic con?guration service) and with their MeGa service (which could be implemented in our architecture with customized Connection classes loaded into the Re?ector). But, differently from them, we focus on providing scalable multimedia distribution using multiple communication protocols, recon?guration of the network topology to deal with failures, and scalable code distribution and dynamic recon?guration. In the last few years, the ?rst commercial Re?ector-like systems for multimedia distribution started to appear (see, for example, VTel's Turbo Cast Re?ector [27] and White Pine's Multipoint Control Unit for Meeting Point [28]). But, unfortunately, literature on this topic is still scarce. Baldi, Picco, and Risso designed a videoconference system based on active networks [29]. Their proposed architecture allows clients to customize their videoconference server (or Re?ector) by uploading mobile Java code. The main difference is that, in their approach, the Re?ector is designed to run in active network routers while ours is designed as a user-level application to be executed on a networked workstation. We chose to implement our system in C11 because our experience shows that Java is not yet ready to provide the high-performance and predictability that most QoSsensitive applications require. Active networks or programmable networks may potentially provide a better throughput and less jitter, since routing decisions are performed at specialized routers rather than on commodity workstations. However, active and programmable networks are not available except for in a few research laboratories. It is not yet clear if these technologies will become widespread in the next years. 8. Conclusions and future work Multimedia applications dealing with large-scale, realtime streaming will become commonplace in future computer environments. The existing hardware and software infrastructure, however, does not provide the high degree of quality of service, ?exibility, con?gurability, and scalability


F. Kon et al. / Computer Communications 24 (2001) 105±123 video, in: WWW Consortium Workshop on Real Time Multimedia and the Web, Sophia Antipolis, 1996. F. Kon, R.H. Campbell, S. Tan, M. Valdez, Z. Chen, J. Wong, A component-based architecture for scalable distributed multimedia, in: Proceedings of the 14th International Conference on Advanced Science and Technology (ICAST'98), Lucent Technologies, Naperville, 1998, pp. 121±135. R. Steinmetz, K. Nahrstedt, Multimedia: Computing, Communications, and Applications, Prentice-Hall, Englewood Cliffs, NJ, 1995. OMG, CORBA v2.2 Speci?cation, Object Management Group, Framingham, MA, OMG Document 98-07-01, February 1998. OMG, CORBAservices: Common Object Services Speci?cation, Object Management Group, Framingham, MA, OMG Document 98-12-09, 1998. S. Deering, Host Extensions for IP Multicasting, RFC 1112, August 1989. H. Schulzrinne, S. Casner, R. Frederick, V. Jacobson, RTP: a transport protocol for real-time applications, revision of RFC 1889, January 2000. P. Sinha, N. Venkitaraman, R. Sivakumar, V. Bharghavan, WTCP: a reliable transport protocol for wireless wide-area networks, in: Proceedings of ACM Mobicom, Seattle, WA, 1999. M.C. Chan, A.A. Lazar, Designing a CORBA-based high performance open programmable signalling system for ATM switching platforms, IEEE Journal on Selected Areas in Communications 17 (9). M.P. Golombek, et al., Overview of the Mars Path?nder Mission and assessment of landing site predictions, Science 278 (5344) (1997) 1734±1742. C.K. Hess, D. Raila, R.H. Campbell, Design and performance of MPEG streaming to palmtop computers, in: Multimedia Computing and Networking 2000 (MMCN00), San Jose, CA, 2000. J. Purtilo, R. Cole, R. Schlichting (Eds.), Fourth International Conference on Con?gurable Distributed Systems, IEEE, 1998. F. Kon, R.H. Campbell, Supporting automatic con?guration of component-based distributed systems, in: Proceedings of the Fifth USENIX Conference on Object-Oriented Technologies and Systems, San Diego, CA, 1999, pp. 175±187. F. Kon, Automatic con?guration of component-based distributed systems, PhD thesis, Department of Computer Science, University of Illinois at Urbana-Champaign, May 2000. S. Frùlund, J. Koistinen, Quality of service aware distributed object systems, in: Proceedings of the Fifth USENIX Conference on ObjectOriented Technology and Systems (COOTS'99), San Diego, 1999, pp. 69±83. J. Nieh, M.S. Lam, The design, implementation and evaluation of SMART: a scheduler for multimedia applications, in: Proceedings of the 16th Symposium on Operating Systems Principles, ACM, Saint Malo, France, 1997. K. Nahrstedt, H.hua Chu, S. Narayan, QoS-aware resource management for distributed multimedia applications, Journal of High-Speed Networking, Special Issue on Multimedia Networking 7 (1998) 227± 255. D. Kotz, R.S. Gray, Mobile agents and the future of the Internet, ACM Operating Systems Review 33 (3) (1999) 7±13. J. Bredin, D. Kotz, D. Rus, Market-based resource control for mobile agents, in: Proceedings of the Second International Conference on Autonomous Agents, 1998, pp. 197±204. I. Goldberg, D. Wagner, R. Thomas, E.A. Brewer, A secure environment for untrusted helper applications: con?ning the Wiley Hacker, in: Proceedings of the USENIX Security Symposium, 1996. F. Kon, B. Gill, M. Anand, R.H. Campbell, M.D. Mickunas, Secure dynamic recon?guration of scalable CORBA systems with mobile agents, in: Proceedings IEEE Symposium on Agent Systems and Applications/Mobile Agents, Zurich, 2000. ? ? F. Kon, M. Roman, P. Liu, J. Mao, T. Yamane, L.C. Magalhaes, R.H. Campbell, Monitoring, security, and dynamic con?guration with the dynamicTAO re?ective ORB, in: Proceedings of the IFIP/ACM

required by these applications. Using our ?ve-year experience with Internet multimedia streaming technology, we have designed and implemented an architecture for scalable multimedia distribution that meets these requirements. In this article, we presented a detailed description of our system and discussed how the lessons learned in early experiments motivated us to implement the enhancements that led to the current architecture. As a future improvement to the Re?ector system, it would be important to develop a service providing ?intelligent? management of the data ?ows among Re?ectors and between Re?ectors and end-user clients. In the current implementation, a bandwidth control subsystem is used to measure the bandwidth utilization for each input and output connection. It could be easily extended to limit the number of users who are allowed to access a particular Re?ector at a given time. With further research, we could also use the bandwidth control subsystem to (1) redirect multimedia streams to different paths in the network to optimize bandwidth utilization and (2) to select which Re?ector should be used by each client based on their location and on Re?ector load. Another interesting future work would be the development of a subclass of Connection, called RSVPConnection, that would reserve network bandwidth using the standard RSVP protocol [30]. This seems to be a relatively easy task if we reuse an existing middleware supporting RSVP such as the one developed by the MONET research group at the University of Illinois. We believe that the architecture presented in this article has a great potential to in?uence the next generation of scalable multimedia distribution systems. But, the methods we describe in this article apply not only to multimedia systems. They give signi?cant insights on how to manage any QoS-sensitive distributed application with respect to scalability, dynamic con?guration, and fault-tolerance. Acknowledgements The authors gratefully acknowledge the work performed by Miguel Valdez, Jim Wong, and See-Mong Tan in the early versions of the Re?ector software. Chuck Thompson and Andrew MacGregor provided valuable system administration support. References
[1] H. Eriksson, MBone: the multicast backbone, Communications of the ACM 37 (8) (1994) 54±60. [2] E. Amir, S. McCanne, H. Zhang, An application-level video gateway, in: Proceedings of ACM Multimedia, San Francisco, CA, 1995. [3] Z. Chen, S.-M. Tan, R.H. Campbell, Y. Li, Real-time video and audio in the world wide web, in: Fourth International WWW Conference, Boston, 1995. [4] S. Tan, R. Campbell, Z. Chen, W. Liao, D.K. Raila, F. Kon, M. Valdez, Adaptation and synchronization in low-bandwidth Internet


[6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16]

[17] [18]



[21] [22] [23] [24]


F. Kon et al. / Computer Communications 24 (2001) 105±123 International Conference on Distributed Systems Platforms and Open Distributed Processing (Middleware `2000), no. 1795 in LNCS. New York, 2000, pp. 121±143. [26] E. Amir, S. McCanne, R. Katz, An active service framework and its application to real-time multimedia transcoding, in: Proceedings of the ACM SIGCOMM Conference, Vancouver, BC, 1998. [27] VTel, TurboCast Architecture, http://www.turbocast.vtel.com, 2000. [28] White Pine, Deploying H.323 Conferencing on Your IP Network,


white paper available at http://www.wpine.com/meetingpoint, 1999. [29] M. Baldi, G.P. Picco, F. Risso, Designing a videoconference system for active networks, in: Proceedings of the Second International Workshop on Mobile Agents (MA `98), 1998, pp. 273±284. [30] R. Braden, et al. Resource ReSerVation Protocol (RSVP) ? Version 1 Functional Speci?cation, IETF Proposed Standard, RFC 2205, September 1997.

Figure 151 A distributed multimedia system_图文.ppt
Figure 151 A distributed multimedia system_调查/报告_表格/模板_实用文档。...distribution to clients Instructor’s Guide for Coulouris, Dollimore and ...
A scalable distributed multimedia directory I.W.Mar....pdf
A scalable distributed multimedia directory I.W.Marshall BT Labs_专业资料。A...(WWW) is the only multimedia distribution and retrieval system which has so...
IMS based Peer-to-Peer content distribution service....pdf
2. New service requirement: How to provide a scalable, reliable and secure multimedia distribution system on FMC IMS for millions of subscribers? e.g. ...
Multimedia Information Release System Operation Man....doc
Multimedia Information Release System Operation Manual5.0_英语考试_外语学习_教育...(3) Due to the default is dynamic distribution and setting static IP, ...
Milos A Multimedia Content Management System for Di....pdf
distribution as fundamental design objectives of the system architecture; (iii...cient, ?exible and scalable multimedia digital libraries. 2 Architecture MILOS...
...and Service Creation for Multimedia Networks.pdf
conceptual model for building scalable multimedia distribution services based on...the process of associating (interconnecting) different components of a system....
...system for on-demand multimedia sessions over In....pdf
multimedia sessions over Internet_工学_高等教育_...[2], have made available a scalable and ...s distribution heterogeneity, it has to be ...
MC_04_Multimedia Basic_图文.ppt
As a means for distribution and presentation of ...Communication system Communicationcapable multimedia ...Graphics is Scalable, which means that the image...
Building a Scalable Multimedia Search Engine Using In?niband Qi Chen1 , ...niband to address the distribution challenge. Basic Design: Figure 3 shows ...
...Video Compression for the Pandora Multimedia System_图文_....pdf
Contents Video Compression for the Pandora Multimedia System_专业资料。Contents...codewords is not adjusted when the frequency distribution of the symbols ...
New Channels for Carrying Copyright and Usage Right....pdf
RIGHTS DATA IN DIGITAL MULTIMEDIA DISTRIBUTION Ahmet...system [27] since this system provides the type...J. Delp, “Protection of Multicast Scalable Video...
MULTIMEDIA INFORMATION DISTRIBUTION SYSTEM 02 睿景多媒体信息发布系统 Realimage ...而且可以在企事业单位的各个区域播放内部信息,通过变换的图文、视频信息,使得办公...
...center power distribution architecture_图文.pdf
a scalable reconfigurable and efficient data center power distribution ...and describes a more effective power distribution system that can be implemented...
...Multi-Source Content Distribution System.pdf
Xeja A Scalable Channel-Based Multi-Source Content Distribution System - Aiming at improving file...
...in a video distribution framework for mobi le en....pdf
quite recently, and the type of multimedia system found embedded in cars. ...cient and scalable on-demand video distribution over wireless networks. The ...
Peer-to-Peer Streaming Systems_图文.ppt
Application Layer Multicast Content Distribution Networks...Advent of multimedia technology and broadband surge...High uplink throughput leads to scalable P2P ...
...Massively Scalable Persistent Content Distribution System_....pdf
Abstract A Massively Scalable Persistent Content Distribution System_专业资料。This paper proposes a novel form of peercasting system as an improved solution ...
3G Wireless Multimedia Technologies and Practical I....pdf
Fig. 1: Wireless streaming multimedia system ...multimedia distribution subsystem, and multimedia ...Bai and S. Wang, “Scalable video streaming ...
Scalable Vector Graphics 可缩放的矢量图形 GIF Graphics Interchange Format 图形...Multimedia database management system 多媒体数据库管理系统 Pervasive 扩大的,...
SAP MM模块培训_图文.ppt
(Sales & Distribution) 产品成本(Product Cost) 工厂维护(Plant Maintenance) ...(Steps to Structure a Classification System T-code:CL00) (1)维护特征 (...

All rights reserved Powered by 甜梦文库 9512.net

copyright ©right 2010-2021。