Ulrich Gall, Franz J. Hauck
IMMD IV, Department of Computer Science
University of Erlangen-Nürnberg, Germany
3SOFT GmbH, Erlangen, Germany
uhgall@informatik.uni-erlangen.de,
hauck@informatik.uni-erlangen.de
The World Wide Web has evolved from a distributed hypertext system to a platform-independent graphical user interface that integrates many network services. So far, its technology has restricted it mainly to applications for information retrieval.Topics: Integration of real-time channels, innovative applications, programmatic and specialized user interfaces
As networks become ubiquitous and more and more users have a permanent connection, there is an increasing demand for other network services, such as real-time data feeds, group communication, and teleconferencing. So far, these services have been provided by various proprietary software systems which were hard to set up and use and thus not very successful.
Integrating real-time group communication services into the World Wide Web is a natural way to make them more accessible and will take the Web a step farther on its way to becoming the universal network application.
In this paper, we describe functionalities required for these services and present an implementation based on Sun Microsystem's Java programming language. We focus on the high-level functionalities and abstractions, but also describe an object-oriented programming model for group communication systems.
Note: If you have a Java-enabled Web browser and a direct connection to the Internet without a fire wall in between, you may want to read the version with live Java applets.
Most people agree that communicating is the most important activity in today's information society. We spend a very significant portion of our time - work and leisure - communicating in groups with other people, either directly (face to face), or remotely (via telephone, email, or videoconferences). Therefore, finding more efficient means for communication between people is a very important area of research. We believe that the most significant problem that keeps people from communicating with one another via teleconferencing software systems is their limited accessibility.
The World Wide Web has become the most convenient way to access information in the Internet. The main reason for this is that WWW browsers integrate different network services into a common, easily accessible, platform-independent user interface.
Developing group communication services that support conferencing from within the WWW is an excellent way to improve accessibility of such services. This should be done in a way that does not require specific proprietary features of WWW browsers; it should be built on general mechanisms provided by all browsers.
So far, the WWW does not support most of the features required for these kinds of applications. For example, there is no mechanism for server-initiated activity. HTTP is not intended for real-time data feeds - it is state- and connectionless. Using CGI scripts and forms, the user interacts with the Web server rather than the browser. Active contents, such as Java Applets, solve these problems. They are a very general mechanism supported by most browsers, yet they are powerful enough to develop serious network-aware applications.
In this paper, we will describe some general design issues for
group communication systems for the WWW and present an implementation
of such a system. The paper is organized as follows.
SectionÊ2 provides a short introduction to
group communication. In SectionÊ3, we discuss
some design issues for the system. We present our proof-of-concept
implementation, called Promondia, in SectionÊ4 and discuss related work in SectionÊ5.
Finally, SectionÊ6 summarizes our results.
2. Group Communication
Group communication is the exchange of information between a group of participants in a sessionÕ. Participants may have different roles in a session. Information is exchanged using certain media, chosen according to availability and suitability for the information to be transferred.
Different types of group communication sessions have evolved in the past in order to reach common goals as efficiently as possible. These goals include entertainment, decision making, learning, and plain information exchange about a common interest.
Figure 1. Session of a panel discussion.
For example, a common pattern is a panel discussion involving a host, several speakers, and an audience (see Figure 1). Typically, all these participants are in the same room. Host and speakers communicate to the audience and among one another using audio and gestures. The audience provides feedback using facial expressions and gestures. The host may let individuals ask questions. The goal is to communicate information the audience is interested in.
This type of group communication is based mostly on the availability of media, not on their suitability. New technology provides the opportunity to choose more appropriate media to reach a given goal more efficiently. Using this new technology in the example above could mean that the audience, distributed all over the world, can provide detailed feedback without disrupting the speakers. This feedback can be summarized automatically, the host and the speakers can see immediate statistics and react accordingly. Anonymous questions can be submitted to the host, who could choose to forward them or discard them.
Group communication can be roughly divided into two categories, synchronous and asynchronous:
| Synchronous | Asynchronous | |
|---|---|---|
| Time frame | "Real-time" Seconds | Delayed Minutes to indefinitely |
| Transfer of information | When available | When requested |
| Analogy in traditional media | Telephony (bidirectional), life TV (unidirectional) | Voice mail, Bill boards |
| New technologies | Videoconference, telephony, shared whiteboard | E-mail, threaded discussion systems |
| Products | ShowMe (SUN Microsystems) ProShare (Intel Inc.) |
Lotus Notes (Lotus) Databases in general |
Asynchronous communication can be viewed as synchronous communication with one virtual participant (like a VCR recording a TV show) that stores information and provides them when requested by another participant. Thus, synchronous communication is the more general concept of the two.
Asynchronous communication can easily be integrated into the WWW using HTML forms or Java front ends and CGI scripts or database servers. Many products and services have been developed for this.
Uni-directional real-time data feeds can also be seen as a special case of group communication with one or more participants, which are simply data sources, and a large number of passive recipients.
In this paper, we focus on real-time, synchronous group
communication and on how it can be integrated into the World Wide
Web.
3. Design Aspects for Group Communication in the
WWW
Most systems for group communication in use today do not clearly separate all of the four components that make up their characteristics: network topology, network services, programming models for and implementations of session types.
There should be a clear distinction between these components. We will discuss the first three in this section and the last one in Section 4.
Certainly, it is helpful to make a clear distinction between a server subsystem and clients when designing an architecture for group collaboration in the WWW. The server subsystem provides functionality common to all session types, such as logging, access control, and directory services. For each session, it also contains a session-server object that coordinates the communication between the clients and determines the contents distributed in that session. This session-server object may be generic or session-type specific.
A client runs on the participant's system and provides the front end that presents the session to the participant. Different types of clients reflect the different roles for each session type; when a participant joins, the appropriate client is instantiated.
Figure 2. Servers with different session-server Objects.
For sessions with a large number of participants (>1000), a scalable architecture is required. This can be achieved using a distributed server subsystem. A network of servers share a common name space and host server proxies on behalf of their clients. These proxies are also session-server objects. They may be generic and simply forward information to and from a central main server, or they may contain semantics for the type of session they serve (e.g., they summarize the local results of a voting session).
Since the the session-server objects may be loaded via the net and instantiated dynamically, a security model is required that restricts their access to local resources. Therefore, Java's coarse-grained security model is not suitable. The more fine-grained security model described in [Rie96] may be appropriate.
The system should be based on a reliable network service to set up the software components and to exchange session and user management data. However, for the actual information to be communicated, it should be possible to use other network services based on the requirements of the particular type of communication. In this case, the system network is used to exchange the references required to use the other network services.
Network services that could be used include TCP, UDP, IP Multicast, and more specialized protocols such as RTP [SCFV96] or TIBCO Inc.'s Rendezvous information bus [TIBC96]. A very early version of our system [Bego96] used IRC [OiRe93] as the messaging backbone. CORBA Events services [OHE96] could be used for strongly typed, language-independent event multicasting.
Many programming models for real-time collaboration have been suggested. The simplest and most basic mechanism is message passing between the objects associated with the participants. A convenient API can offer asynchronous multicast for remote method invocation on all or a subset of participants [RoGr96]. This model is very suitable and intuitive to use for volatile communication-oriented tasks, such as chatting. It can be refined by distinguishing certain characteristics for the messages (e.g., unreliable delivery or unspecified delivery order).
To identify different types of messages, it has proven practical to allow for more than one named communication channel per session [ITU122]. Clients interested in a particular type of message can subscribe to its respective channel. That way, the different clients reflecting the different roles only receive the information they need.
To allow for a reply to the messages sent, delayed return values for asynchronous method invocation can be implemented by creating a handle to represent the reply expected in the future. This reference can then be used to wait for the reply or to forward it somewhere. An extension of Java providing this is E, described in [E96].
Other, more sophisticated models can be built on top of this simple infrastructure.
Model-View-Controller-based (MVC) approaches with a replicated model, one view and one controller per user are very suitable for collaboratively creating or modifying data. However, they tend to be inefficient if all changes to the model are immediately sent to all views. However, Graham et al. report on significant performance advantages using optimized network communication [GUN96]. If a more sophisticated strategy is used to transmit changes only when needed, application specific semantics have to be considered. This can be quite difficult to implement.
Gutfreund et al. suggest a Linda-like simplified tuple space model for coordinating distributed applications [WWWi96, Gele89]. This is a very lightweight approach to the MVC pattern that provides a generic server and requires all semantics in the client objects. This can also be used for simple message passing, since clients may register for changes on tuples. Lotus has a similar approach, using "things" as state-keeping objects that also provide notification services [DPK+96].
For synchronisation and floor control, token-based mechanisms can be used to ensure data integrity [Burr96] . This is an easy way to use collaboration-unaware programs collaboratively, but severely restricts interaction between participants.
Finally, real-time multimedia streaming requires sophisticated network protocols with different quality of service characteristics. It should be possible to integrate these services and have them controlled by the application.
All these different approaches are very suitable for certain tasks, whereas they are inconvenient or inefficient for others. We believe that these models should complement each other rather than compete. A framework for group communication should not be bound to a particular model, but provide different models for the developer to choose from.
To allow for software reusability, it should be possible to combine different session types to form a new type. This can be done recursively to create a hierarchy of components and to dynamically add new ways of communication to a session when required. When a client joins a session containing a subsession, it automatically joins the subsessions available in that session, too.
For integrating group communication into the Web, a session should run within a Web page using standard Web browsers. That way, they can be combined with HTML data containing general information about the session or its topic. Also, sessions can be embedded into WWW pages about a particular subject to let people, interested in this subject, communicate with each other and create a virtual community.
Standard WWW mechanisms can be used for directory services and
bookmarking. Sessions can be embedded into MIME email messages so that
they are automatically started when the recipient reads the email
message. This is a convenient way to send invitations for sessions.
Within a session, Web pages can be shown to all participants to point
to information relevant for the current topic.
4. Promondia: A Proof-of-concept
Implementation
We have implemented a system, called Promondia, providing group communication functionality for the WWW. An early version of our system is known under the name COMO.
Promondia consists of a server program and session starters implemented as Java applets, which are embedded in HTML documents. These applets are called session-management applets.
When the user views an HTML document that contains a reference to a session-management applet, the browser loads this applet via HTTP and starts it. The applet does not necessarily have to be located on the same WWW server as the HTML page containing it. However, access via certain WWW servers can be restricted.
When the applet is started, it connects to the Promondia server and requests to join the specified session. If that session does not exist, a new session is created on the server by instantiating the specified session-server object. When a session is joined, the session-starter applet instantiates the session client reflecting the participant's role in the session.
The Promondia server is a stand-alone Java application. It is entirely written in Java and does not require any native code. Thus, it runs on any platform for which a Java interpreter is available.
The server includes an HTTP server. This is very convenient for the following reason: Current Web browsers allow applets to connect only to the host they were loaded from. Otherwise, it would be possible to bypass a firewall with a malicious applet. Thus, the Java applets must be loaded from the system the Promondia server is running on. This can be achieved by using the built-in HTTP server. As an alternative, a separate standard HTTP server can be set up on the same machine.
A group-management module organizes and coordinates the group-communication channels. It also provides information about active sessions. It supports access control on a general, a per-session-type, and a per-session basis. Configuration can be done remotely using special Java applets.
The multithreaded architecture makes up for performance weaknesses that are due to the fact that the Java byte code has to be interpreted. Without just-in-time compilation, a 133 Mhz Pentium PC can handle about 100 simultaneous connections. A Sun UltraSparc workstation can handle several hundred connections at the same time. Using Kaffe [Kaffe96], a freely available runtime system for Java, 350 simultaneous connections have been handled easily on a Pentium 120 running Linux.
Two session management applets have been implemented to embed sessions into an HTML document:
Other session management modules have been implemented as sessions. That way, they can use the communication mechanisms provided by the system to distribute information about available users and session.
A few session managers and communication methods have been implemented to demonstrate the system.
A Manager Session provides information about which sessions are running and which users are available. This is an example of how a session can be used for multicast distribution of information.
It provides two roles: user and administrator. A user can select sessions and join them, which will download the client code for the session and spawn a new window. Invitations can be sent via email; the recipient can join the session simply by reading the email with a Java-enabled browser. The administrator can change the properties and configuration of the sessions and expel users from a session. Fig. 3 shows a screenshot of a Manager-Session client.
Figure 3. The Manager-Session client for administrators.
The Chat system offers a variety of chat modules: a simple one and several more sophisticated ones, which allow for private and anonymous messages, notification at certain events, and audio messages. Users can direct other participants to Web pages by typing a URL. Fig. 4 shows a Chat client with an ongoing session. A Moderator client lets the administrator control what is being said in the session by assigning different rights (free speech, moderated, muted) on how to take part in the conversation.
Promondia Chat sessions can be connected to IRC channels. The information transferred is mapped both ways. If the session is moderated, the IRC channel will also be moderated.
Figure 4. A Chat client.
The Shared Whiteboard is a multi-user vector-oriented drawing program. Lines, polygons, rectangles, ovals, text, and imported images can be added to a shared canvas (see Fig. 5). This kind of functionality can be used as a subsession of a conferencing system to illustrate ideas or discuss designs.
Figure 5. The Shared Whiteboard client with a loaded image and some
graphics.
A VoteModerator applet lets an administrator configure a question and a set of suggested answers. The users can choose one of the answers to cast their votes using a Voter client. The results can be transmitted automatically or after approval by the moderator. In Fig. 6, a Voter is shown. The already known votes are presented as colored bars behind the suggested answers.
A
simple board game for 2 to 10 players has also been implemented to
illustrate that multi-player games can be seen as a form of group
communication with very formal interaction rules.
Most work on Web-based collaboration focuses on asynchronous
database- and transaction-oriented cooperation. An example is the
BSCW system [BHST95, ApBu96] developed as part of the CoopWWW
project [Appe96]. Current implementations of
synchronous communication for the Web are mostly limited to chats or
other predefined functionality.
GroubWeb [GrRo96] is a collaborative
Web Browser that is built on top of GroupKit [RoGr96]. This approach can even make Web browsing a
fully collaborative activity; however it does not improve accessibility
of group communication, since proprietary software has to be
installed. The same applies to RAVE [EAU95], which uses browser
plug-ins for presenting real-time data feeds. CoWeb [JGKZ95]
solves these problems by using Java, but it is not programmable.
GroCo is a simple, but flexible system based on message passing and
implemented for the alpha release of Java [Walt95, Walt96]. The data
objects sent in the messages are restricted to certain types.
Mushroom is a very promising and ambitious project that
focuses on an intuitive user interface. A room metaphor is used
to represent a session. Persistent data can be stored in a room for
asynchronous cooperation. Users may enter a room and meet for
synchronous cooperation. [KCDH96, Kind96]. A Java-based implementation is under way,
but not available yet.
JavaSoft is working on a collaboration API [Burr96] that is based on ITU recommendation T.122
[ITU122]. This API does not yet cover
higher-level functionality for session management and configuration.
We have discussed some of the issues involved in designing an
architecture for a group communication infrastructure for the World
Wide Web. We believe that building such an infrastructure can help
people to communicate more efficiently, since it would make these
application more accessible.
Promondia, our award-winningÓ
implementation of such a system is freely available for academic and
non-profit organizations at
http://www6.informatik.uni-erlangen.de/promondia/. It is used to
provide chat services on many public Web sitesÒ
in the Internet. It is unique in that it is the only flexible,
high level infrastructure currently available for building synchronous
group-communication applications for the Web. However, it lacks many of
the features necessary for a globally usable system.
There is still a lot of work to be done. In particular, we are
currently working on a scalable reliable multicast backbone. The
programming models that have been suggested in the past can be combined
to let the developer choose the model that is most suitable for the
task. However, we feel that it would be helpful to integrate these
models in a way that allows for component reusability. For that, we are
currently investigating the usability of a model with a hierarchy of
light weight communication channels and with a very weak distinction
between channels and sessions.
[ApBu96]
W. Appelt, U. Busbach: "The BSCW system: a WWW based application to
support cooperation of distributed groups." 1996.
[Appe96]
W. Appelt: "CoopWWW: interoperable tools for cooperation support using
the World Wide Web." In: Proc. of the ERCIM Workshop on CSCW and the
Web (St. Augustin, Germany, Feb. 1996).
[Bego96]
Begole, J. (1996): Review of Como
[BHST95]
R. Bentley, T. Horstmann, K. Sikkel, J. Trevor: "Supporting collaborative
information sharing with the World Wide Web: the BSCW shared workspace system."
In: Proc. of the 4th WWW Conf. (Boston, MA, 1995).
[Burr96]
R. Burridge: Java Shared Data API.
[DPK+96]
M. Day, J. Patterson, J. Kucan, W. Meng Chee, D. Mitchell: Notification
Service Transfer Protocol (NSTP), Version 1.0. Techn. Report 96-08,
Lotus Inc., 1996.
[E96]
Electric Communities: The E programmer's manual, 1996.
[EAU95]
P. England, R. Allen, R. Underwood: "RAVE: Real-time services for the Web."
In: Computer Networks and ISDN Systems, Volume 28, pp. 1547-1558, Elsevier, May 1996.
[Gele89]
D. Gelernter. "Multiple tuple spaces in Linda." In: Proc. of PARLE '89,
Springer-Verlag, 1989, pp. 20-27.
[GrRo96]
S. Greenberg, M. Roseman: "GroupWeb: A Groupware Web Browser."
In: Video Proceedings of ACM CSCW'96 Conference on Computer Supported Cooperative Work.
Figure 6. A Voter client.
4.3.5 Games
5. Related Work
6. Conclusion and Future Work
References
[GUN96]
T.C.N. Graham, T. Urnes, R. Nejabi: "Efficient distributed implementation
of semi-replicated synchronous groupware." In: Proc. of the ACM Symp. on User
Interf. Softw. and Techn., ACM Press, Nov. 1996
[ITU122]
International Telecommunication Union Recommendation T.122
[JGRZ95]
S. Jacobs, M. Gebhardt, S. Kethers, W. Rzasa: "Filling HTML forms simultaneously: CoWeb -
architecture and funcitonality"
In: Computer Networks and ISDN Systems, Volume 28, pp. 1385-1395, Elsevier, May 1996.
[Kaffe96]
Kaffe Home Page. 1996.
[KCDH96] T. Kindberg, G. Coulouris, J. Dollimore, J. Heikkinen: "Sharing objects over the Internet: the Mushroom approach." In: Proc. of Global Internet 96 (London, UK, Nov. 1996).
[Kind96]
T. Kindberg: "Mushroom: a framework for collaboration and interaction
across the Internet." In: Proc. of the ERCIM Workshop on CSCW and the
Web (St. Augustin, Germany, Feb. 1996).
[ORF96] Robert Orfali, Dan Harkey, Jeri Edwards: The Essential Distributed Objects Survivial Guide, John Wiley & Sons Inc.,New York 1996
[OiRe93]
J. Oikarinen, D. Reed: Internet Relay Chat Protocol. Request for
Comments #1459, May 1993.
[Rie96]
T. Riechmann: Security in large distributed, object-oriented systems.
Techn. Report TR-I4-96-02, IMMD IV, Univ. of Erlangen-NŸrnberg, June 1996.
[RoGr96]
M. Roseman, S. Greenberg: "Building real time groupware with GroupKit,
a groupware toolkit." In: ACM Trans. on Computer Human Interaction,
ACM Press, March 1996.
[SCFV96]
H. Schulzrinne, S. Casner, R. Frederick, V. Jacobson: RTP: a transport
protocol for real-time applications. Request for Comments #1889, Jan. 1996.
[TIBC96]
TIBCO Inc.: White Paper [about the Rendevouz Information Bus], 1996.
[Walt95] M. Walther: Primitives for group communications in electronic meeting systems. Diss., Bond University, Canberra, Australia, 1995.
[Walt96]
M. Walther: "Supporting development of synchronous collaboration tools
on the Web with GroCo". In: Proc. of the ERCIM Workshop on CSCW and the
Web (St. Augustin, Germany, Feb. 1996).
[WWWi96]
WWWinda Home Page.