Videoconferencing in heterogeneous environments

An Integrated Videoconferencing System
for Heterogeneous Multimedia Collaboration
Ahmet Uyar1,2, Wenjun Wu2, Hasan Bulut2, Geoffrey Fox2
1
Department of Electrical Engineering and Computer Science, Syracuse University
2
Community Grids Lab, Indiana University
auyar@syr.edu, wewu@indiana.edu, hbulut@indiana.edu, gcf@indiana.edu

Abstract range of collaboration functions and interacts easily with
these systems.
We have developed an integrated conferencing Although there has been some work done to
system, Global Multimedia Collaboration System, which interoperate these systems, they exclusively focused on
enables heterogeneous multimedia clients to join the the interactions between two communities. [4,5]
same real-time sessions. Our system provides support presents a solution to interoperate SIP and H.323
for a variety of protocols and applications, including systems. [6] is another videoconferencing system which
H.323 clients, SIP clients and Access Grid rooms. In targets the interoperability between H.323 and Access
this paper, we would like to show the features and Grid communities. Our approach is more general and
design principles of our conferencing server which provides a general session management protocol to
facilitates audio and video communications among accommodate more protocols and applications with an
participating clients in a real-time conference. easy to use web interface. It is even not limited to
multimedia sessions, it can be extended to be used for
Keywords: videoconferencing, XGSP, Collaborative any kind of real time application from simple chat to
Systems and Applications, Distributed Multimedia online gaming.
Systems. In this paper after giving a brief overview of our
Global-MMCS, we will provide a detailed description of
our audio and video conferencing solutions and their
1. Introduction design principals, and discuss the performance results.

Today Internet provides a very convenient and cost 2. Global-MMCS Overview
effective medium for audio and video communications.
Highly sophisticated audio and video codecs have been
designed to transfer audio and video content efficiently Global-MMCS[7,8,9] is an integrated video
and standards have been developed to initiate and conferencing solution which enables heterogeneous
manage real-time multimedia sessions. However, these clients to join the same real-time multimedia sessions. It
standards come from different communities and they provides a flexible architecture to support even more
tend to solve the same problem in quite different ways, standards and applications. There are five main
partly because of some historical reasons and partly components of this architecture (Figure 1);
because of differences on objectives. Currently there are NaradaBrokering(NB) servers, XGSP Session Server,
three major video conferencing systems, H.323[1], Gateways, Media Server, and Web server. NB is a
SIP[2] and Access Grid[3]. Every community has its distributed publish/subscribe messaging system which
own set of protocols and products. A user of one system delivers all messages. XGSP Session Server manages
can not talk to a user of another system, although both real-time sessions, namely starts/stops/modifies them. It
have the means to send/receive audio and video. They receives messages from gateways and the web server,
are like three islands in Internet without a bridge among and performs appropriate actions on the media server.
them. Gateways receive protocol specific messages from
We have designed an architecture and implemented different clients, and pass them to the XGSP session
the prototype system to integrate all these different server after converting to XGSP messages. There should
communities in the same real-time session in an easy-to- be a dedicated gateway for each supported protocol.
use fashion. Since these systems are not identical in Currently we have H.323 and SIP gateways. Media
functionality and their way of solving the same problem Server facilitates the audio and video communications
is quite different, it is not easy to translate messages among participants in a meeting. We will cover the
from one protocol to another. Therefore, we have details of the media server in following sections.
developed a more general XML based session The web server provides an easy-to-use web
management protocol (XGSP) which covers a wide interface for users to join multimedia sessions and for
administrators to perform administrative tasks. In
addition, users can start some audio and video clients

through these web pages such as VIC, RAT and Real for developing multimedia applications, since it
Player. provides the performance of the native code and the
flexibility of Java platform. By default it supports the
Media Servers most commonly used codecs such as H.261, H.263,
(Audio, Video, Image JPEG video codecs and G.711, G.723.1, GSM audio
XGSP Grabber Servers) codecs. All these codecs come with both encoders and
Web
Session decoders except H.261. It only has the decoder. This
Server
Server
was an important missing part for us and we have
developed an encoder for H.261.
NaradaBrokering Today most of the conferencing servers are based on
All Messaging
H.323. There are hardware and software
implementations. Polycom and Radvision are two
companies which provide hardware based conferencing
SIP H.323 servers. IBM Lotus Sametime, VCON Media Exchange
Gateway Gateway Manager, and OpenMCU H.323 conference server are
some of the software based H.323 conferencing servers.
There are also some projects[11] to implement SIP
SIP H.323
Client
based conferencing servers.
Client

4. Audio Conferencing
Figure 1. The architecture of Global Multimedia
Collaboration System Contrary to human eyes, human ears are very
sensitive to distortions in voice. In addition, human ears
are also sensitive to the delay of voice from a speaker to
3. Media Server
the listener. Studies show that to give the feeling of one
is having a direct conversation with the other party, the
Media server facilitates audio and video transfer delay in the transmission of audio should not exceed
among participants in real-time multimedia sessions. It 400ms. Therefore we need to pay close attention to
receives RTP streams from source clients and these factors when designing an audio conferencing
redistribute it to interested parties, sometimes by system. The delay is introduced by a combination of
replicating the received media, sometimes by mixing, factors, but the most important ones are the delays
and sometimes by transcoding. In addition, it provides caused by transmission, buffering and mixing. The
an XML based message interface through which audio distortion in the voice can be caused either by the
and video sessions are managed by the session server. It package loss during the transmission or by the
has three main components; audio server, video server accidental deletion of speech data when suppressing the
and image grabber server. Each of these three silence. Although the loss in the transmission is not in
components are independent of one another and they run the scope of this paper, silence suppression is a very
in different machines. This makes our system more important part of any audio conferencing solution.
scalable and flexible. Audio conferencing over Internet can be
Both Audio and Video Servers implement a meeting implemented in many ways[11]. Here we give four
management concept and an interface to manage these different architectures.
meetings over the network. Session Server manages A. Endpoint Mixing: All audio streams are
these meetings by sending/receiving messages back and delivered from audio sources to recipients without
forth over the network. They provide functions to modifying or mixing them in server side. Audio
create/delete meetings, and to add/remove participants. middleware redistributes audio packages to interested
In addition, they also provide a method to give the list of parties by replicating them whenever necessary. A client
audio and video streams in meetings. These are all XML may get more than one audio stream at a time and it
messages and defined in an XML schema. Both Audio should be able to mix and play them. IP multicast or
and Video servers support many meetings at a time and software-based multicast audio conferencing
each meeting can have any number of participants. solutions[12] use this mechanism. The advantage of this
We have implemented the media server in Java solution is that the audio middleware introduces
using Java Media Framework[10]. It provides a flexible minimal transmission delay since there is no mixing or
and extendible architecture. Since Java is relatively transcoding during transmission. In addition, this
slower than C/C++ for multimedia processing, all CPU solution scales very well, since the only computing
intensive encoders and decoders are implemented in performed in server side is to replicate the data. On the
native C/C++ code. This makes JMF a good candidate

other hand, this solution requires more bandwidth for mixed stream. The main disadvantage of this solution is
each client since they may receive more than one audio its scalability. Since there is only one mixer in the
stream at a time. Moreover, for this solution to work system, it can not support thousands of clients in a
efficiently, each audio sender client should implement a session. Nonetheless, our experiments show that such a
silence suppression mechanism so that when a system can support up to 300 clients easily.
participant is not speaking, no audio data should be sent
from that client. Otherwise, the scalability of this system 4.1 Audio Server Implementation
will be severely affected and bandwidth sensitive clients
will end up loosing some valuable audio data. We have chosen the centralized audio mixing model
Furthermore, each client should support all the codecs (Figure 2) from the listed alternative conferencing
used in a session by all participants, since there is no architectures above. It is easy to implement and flexible
transcoding during transmission. enough to support variety of clients with different
B. Distributed Mixing: Another way of capabilities. Its scalability is also good enough as it can
implementing audio conferencing over Internet is to be seen from performance figures.
move audio mixing process from audio endpoints to the One important point we need to consider while
server. This architecture will be very similar to the designing an audio server is the silence suppression.
previous one except it will handle audio mixing on the Although it is best to suppress the silence in audio
server. There will be one audio mixer for each streams at the source clients before sending it over the
participant in a meeting and each participant will receive network, we assume that the incoming streams may not
exactly one audio stream. This mechanism reduces the be silence suppressed. Because we do not have any
bandwidth requirement for the end user significantly. In control over audio clients. Therefore, it is essential for
addition, users can be given an option to choose the us to suppress the silence in each incoming audio stream
audio format they want. Moreover, audio streams can be in the server. Silence suppression is particularly
silence suppressed before mixing. Therefore this important because when audio streams are mixed,
architecture can be more flexible to deploy than silence packages add up and produce a lot of unwanted
previous one and a diverse set of audio clients can be noise. In addition, silence suppression saves CPU time
supported with different network bandwidth capabilities. on the server machine when mixing audio packages,
The scalability of this architecture will be similar to the since the silence packages will not get mixed. Moreover,
previous case, but nonetheless it is a challenging task to most of the time there is only one speaker in a meeting,
develop such a distributed audio conferencing solution. and we can avoid mixing the audio streams if we
It is particularly difficult to distribute computing load suppress the silence properly.
among different machines and to maintain such a Various silence detection algorithms proposed with
complicated system Another disadvantage of this different computational complexity and accuracy; HAM
solution is that the mixer in the server will introduce algorithm, Exponential Algorithm, Absolute Algorithm,
some delay to the transmission of packages. However, Differential Algorithm [13], zero crossing rate
our experiments show that if there is only one audio algorithm[14], refined block oriented algorithm [15],
mixer along the way from audio sources to the target, Silence Compression Scheme of G.723.1 [16]. We use
the delay introduced by the mixer can be tolerable. the refined block oriented algorithm for its simplicity
C. Hybrid Model: One can also imagine a hybrid and reasonable accuracy. This algorithm compares the
architecture in which some clients will have a mixer average energy of each package to a threshold and
dedicated for them on the server and some will receive decides that package as silence if enough consecutive
all audio streams from servers directly and handle by packages are below the threshold. Since we do not
themselves. This architecture will have the advantages employ a comfort noise generator, we wait half a second
of both systems and provide services according to the before deciding a package as silence to avoid deleting
capabilities of the end user. Nonetheless this is also a the silence packages between the words of a speaker. In
complicated system and not very easy to develop and addition, to avoid missing the beginning of a speech we
maintain. examine the sub blocks of each package and decide it as
D. Central Mixing: In this model there will be one speech if the average energy of a sub block is higher
central audio mixer in the server and all participating than the threshold.
clients send their audio to that mixer. This mixer will Figure 2 shows how the audio server works. First
mix all audio streams and then send the mixed audio to audio streams are decoded to raw data, then repackatizer
each participant. This solution is less computationally adjusts the sizes of audio packages if necessary. Since
intensive compared to the second approach, since there our system supports a variety of clients, not all of them
is only one audio mixer rather than N audio mixers in a use the same package size. While Polycom client uses
meeting. It also has the same advantages as the second 60ms packages, Rat 4.2.2 uses 20ms packages.
approach, since it is providing each participant with one Currently we use 60ms as our systems package size.

Silence detector passes the speech packages to the memory, Windows XP machine. All machines involved
package queue and packages wait in this queue to be in this test run in a 100Mbps subnet. We tested it with
picked up by the audio mixer. Packages are buffered in one active speaker and two active speakers at a time.
this queue for a while to avoid missing the late arriving Speakers sent a 64kbps ULAW audio stream. When
packages because of the jitter in the transmission time. there is only one active speaker, no mixing is performed.
Audio mixer polls all queues regularly and passes a This is the most common case in video conferencing
copy of mixed audio data to subtracters. Mixer just adds environments since most of the time only one speaker
the values of all available data and store the result in a talks. As it can be seen from Table 1, in this case, it
short array instead of byte to avoid overflow or provides a good quality audio for up to 300 participants.
underflow. Then subtracters subtract the data of After that it starts dropping packages since it can not
themselves from the received mixed data if there is any, process it on time. Although we did not include the
store the result in a byte array and pass it to the encoder. results for two active speakers case, it provides good
If the mixed audio sample value is out of range for byte quality audio for up to 275 participants and then it starts
type, the maximum or the minimum byte value is dropping packages. In this case two active speakers
assigned accordingly. We have not experienced any were always sending audio, so there have been
distortion of audio because of this value conversion. continuous mixing activity.
We have also tested the effects of having multiple
Stream 1
sessions at a time. We created audio sessions with 50
D R SD PQ participants and two active speakers each. Our tests
Stream 2 S E show that audio server can support 5 concurrent sessions
D R SD PQ To
Multicast (250 participants in total) without any package
Group
Multicast Group Streams
Audio droppings. Since there are more mixers and more
Unicast
Mixer To
incoming audio streams, the supported users are a little
User 1
D R SD PQ S E Unicast less than the previous one session cases.
User 1

Unicast To Number CPU Mem. Total Quality
User N PQ Unicast of users Usage Usage BW
D R SD S E
User N
D: Decoder SD: Silence Detector S: Subtracter
(MB) (Mbps)
R: Repacketizer PQ: Package Queue E: Encoder 50 %3 39 3.2 good
100 % 12 60 6.4 good
Figure 2. Centralized Audio Mixing 150 % 24 80 9.9 good
200 % 38 101 12.8 good
Since our system is designed to bridge Access Grid 250 % 49 121 16.0 good
and other videoconferencing systems, we also support 300 % 56 141 19.2 fair
multicast groups in our audio server. It is similar to 350 % 56 161 22.4 poor
unicast support, but in this case usually multiple audio
streams are received. The handling of these streams are Table 1. Audio Server performance results for one
the same up to the mixer, but after the mixer the audio session with one active speaker
subtracter subtracts the values of all streams from that We should note the fact that even when the CPU
multicast address. Therefore it sends the mixed audio of utilization is quite high such as %50, audio server can
all unicast clients to the multicast group in a session. still deliver a good quality audio because mixer runs
We expect our system to work with any H.323 or regularly and CPU is utilized linearly. As long as it
SIP based multimedia client which supports any of the finishes processing a package during the given time
audio codecs, G.711, G.723.1, GSM. We have tested our interval, it will not delay the processing of the next
audio server using an H.323 based Polycom client, SIP package. But if it can not finish it on time, then it will
based HearMe and Windows Messenger clients and delay processing of the next package. Since next
RAT from MBone tools. It works well with all these package also will not be finished on time, it will delay
clients. the processing of the one after that. This will cause
regular package drops from the package queue causing
4.2 Audio Server Performance Tests audio quality degradation for the end user.
The performance of the audio server depends on the
number of participants and active speakers in a session, 5. Video Conferencing
and the number of audio sessions at a time. First we test
the scalability of our audio server for one audio session. The requirements for video transmission are
Audio server runs in a 2.5GHz Pentium 4 CPU, 512MB significantly different than audio transmission. Video

streams entail much more network resources, and its streams, and to mix up to four video streams into one
encoding/decoding requires much more computing and send out to registered destinations. Every session in
power. Moreover, there is no trivial way of mixing video server has a multicast group address, a video
video streams as mixing audio. Although one can merge mixer and one or more unicast users. Each unicast video
a number of streams into one, it is not easy to merge an stream is forwarded to the multicast address and through
arbitrary number of video streams into one. this multicast address image grabber server receives all
Video conferencing can be implemented in many video streams in that session.
different ways. Here we give three different
architectures and discuss the advantages and Unicast clients
disadvantages of them:
A. Multicast Style: In multicast style conferencing,
all participants send video streams to a group address
and network or middleware servers deliver these streams Video Access Grid
to participants by replicating the data whenever Server Multicast Session
necessary. This approach can be implemented either
using IP multicast supporting routers or software based
multicast implementations such as a distributed Image
Grabber
brokering system[12]. Access Grid uses ip-multicast for Server
video conferencing. The advantage of this solution is its
simplicity and ease of use for the end user. The main Figure 3. Video and Image Grabber servers
disadvantage of it is that it requires much more network
bandwidth than many users can accommodate, since all Image grabber server is independent of the video
video streams are delivered to all participants. Even if server and runs in another machine. Because it processes
the network bandwidth is available, the computers the incoming video streams continuously, it requires a
which receive these video streams may not have the lot of computing resources. It first decodes the received
means of processing them. video streams into YUV format and then encodes them
B. Multicast Gateways: This approach is similar to into JPEG format, and regularly saves the snapshots of
first one, but for those users who do not have multicast these streams in JPEG file format. Web server accesses
support or who do not have enough bandwidth, or who these pictures and provides them to users through a web
do not have the capability of processing multiple interface. These pictures gives the end user a pretty
streams, gateway(s)[17] can be placed in the edge of a good sense of what each stream is about, before
multicast network. These gateways receive video deciding to choose a video stream to receive.
streams from multicast network and forward the Figure 4 shows the video mixing algorithm in video
requested video streams to proper destinations. In server. Because of the space limitations, the details of
addition, these gateways can also provide video merging video streams 2 and 3 are not shown but they are
and transcoding services. Moreover, they can implement processed in the same way as the other two streams.
a session management protocol to give users an option After receiving a to be mixed video stream, a replicator
to choose the video stream(s) they want. H.323, SIP or a filter duplicates video packages and passes one copy to
proprietary protocol can be used for this purpose. [18] RTP Transmitter which sends out to proper destinations
uses gateways to merge many video streams into one and another copy to a video decoder which decodes the
and requires users less bandwidth and less computing received stream into YUV format. Video Mixer first
power. RTP
Replicator
RTP
Receiver Transmitter
C. All Unicast: Third approach would be to avoid
using multicast altogether and route the video streams
Video
through software or hardware based servers. The main Decoder
advantage of this solution would be not requiring
Video Stream 2
multicast support and can run anywhere. Although some Video Video RTP
sophisticated servers can handle very large scale Video Stream 3
Mixer Encoder Transmitter

meetings, this solution can be used effectively for small
scale sessions. Video
Decoder

5.1 Video and Image Grabber Server
RTP RTP
Implementations Receiver
Replicator
Transmitter

Our video server(Figure 3) is effectively a multicast
gateway with a general session management interface. Figure 4. Video Mixing in Video Server
Its main functions are to redistribute the received video

reduces the size of these incoming four video streams to In forwarding case, one user sent a H.263 video
one fourth of their original size. Then it combines these stream to the server machine -1.2GHz Intel Pentium III
streams into one full picture. It places every stream to dual CPU, 1GB MEM, RedHat Linux 7.3, and server
one corner of the merged video picture(Figure 5). In the forwarded it to many clients. Instead of measuring the
next step, this newly mixed video stream is encoded CPU load of the server, we have calculated the delay,
either into H.261 or H.263 format and then it is jitter and loss rates for each package over a period of
transmitted to proper destinations by the RTP time. This way we get a more precise way of assessing
transmitter. Since Video mixing is a CPU intensive the perceived performance of our system by users.
application, instead of giving each user an option to Although video server distributed the packages to
create his/her own mixed video stream, we create only hundreds of destinations, we gathered the results from
one video mixer for each session and the system 12 clients for the ease of testing. The sender client and
administrator has the right to choose the video streams 12 receiver clients, from whom we gather results, were
to be mixed. Our web server provides an interface to running on a 2.4GHz Intel Pentium 4 CPU, 512GB
add/remove video streams to/from the mixer. MEM, RedHat Linux 7.3 machine. The video stream
Figure 5 shows the mixed video in VIC, Real Player had an average bandwidth of more than 600 kbps. The
and Polycom windows. In background a snapshot of our sender application sends 2000 packages in each test, 23
web page is also seen. Mixed video streams are second part of a movie. We used the same video stream
particularly important for users who can not receive for each test. All machines involved in this test reside on
more than one video stream such as a Polycom client. a gigabit subnet. For every package we calculated the
These users get the pictures of four participants in a transit delay, (receivedTime – sentTime), for all 12
meeting through one video stream. clients and then we get the average of these 12 delay
values in milliseconds. We also calculate the average
jitter for each package based on the formula explained in
RTP RFC [19].

Number Avg. Avg. Loss Total
of Delay Jitter rate bandwidth
clients (ms) (Mbps)
50 3.08 1.10 0.0% 30
100 10.72 3.34 0.0% 60
200 27.69 7.56 0.4% 120
300 60.86 11.84 0.7% 180
400 229.2 15.55 6.0% 240

Table 2. Video forwarding performance results
Table 2 shows that our video server is capable of
supporting 300 clients if there is only one video sender.
When it sends to 400 clients, it starts dropping packages
Figure 5. Mixed video streams in various windows and also latency becomes pretty high. We should also
Since our system supports H.261 and H.263 codecs note the fact that this video stream had an average
we expect it to work with any H.323 or SIP compatible bandwidth of 600kbps which is quite high. In a normal
multimedia client which supports any one of these video conferencing setting, the average bandwidth of a
codecs. We have tested it with a H.323 based Polycom video stream is much lower.
client and VIC MBone tool. It works well with both of In mixing test, we ran the video mixer in a 1.2GHz
them. Intel Pentium III dual CPU, 1GB MEM, RedHat Linux
7.3 machine. The video mixer mixed four identical
5.2 Video Server Performance Tests H.261 video streams. The bandwidth of this stream
The testing of video server is more complicated changed from 100kbps to 200kbps. In this case, since it
since there are some video streams which are only is not easy to measure the latencies and jitters for each
forwarded and there are also some streams which are video package, we measured the CPU load on the
mixed. We have tested the performance of forwarding server.
and mixing separately. Table 3 shows that video mixing is a CPU intensive
process and this machine can only handle three video

mixers at a time. These results also suggest that video 2002 International Multiconference in Computer
mixing should be done in a separate machine than video Science and Computer Engineering, Internet
forwarding. Although currently these two processes are Computing(IC’02), June 2002, Las Vegas
running in the same machine, we plan to separate these [8] Wenjun Wu, Ahmet Uyar, Hasan Bulut, Geoffrey
functions and distribute them among more servers. Fox, Integration of SIP VoIP and Messaging
. Systems with AccessGrid and H.323, (to appear)
Number of CPU load the proceedings of The 2003 International
Mixers Conference on Web Services (ICWS'03), June
1 15-25% 2003, Las Vegas, ND, USA.
2 30-50% [9] Geoffrey Fox, Wenjun Wu, Ahmet Uyar, Hasan
3 50-70% Bulut, Shrideep Pallickara, Global Multimedia
Collaboration System, 1st International Workshop
Table 3. Video mixing performance results on Middleware for Grid Computing, June 2003, Rio
de Janeiro, Brazil.
[10] Sun Microsystems, Java Media Framework 2.1,
6. Conclusion and Future Work
http://java.sun.com/products/java-
media/jmf/2.1.1/index.html, 2001
In this paper we have presented our conferencing [11] K. Singh, G. Nair, and H. Schulzrinne. “Centralized
system which provides services to a diverse set of conferencing using SIP.” In Internet Telephony
clients. We have also given a detailed description of the Workshop 2001, New York, Apr. 2001.
architectures of our servers and performance results. [12] Ahme Uyar, Shrideep Pallickara, Geoffrey Fox,
The performance results show that our current “Towards an Architecture for Audio/Video
implementation supports hundreds of participants in Conferencing in Distributed Brokering Systems”,
audio and video sessions. It also shows that JMF can be To appear in the proceedings of The 2003
used to implement such conferencing systems. International Conference on Communications in
In the next step, we are planning to develop a Computing, June 23 - 26, Las Vegas, Nevada,
distributed conferencing system which will be USA.
implemented on top of a distributed brokering system. [13] Claypool M., Riedl J., “Silence Is Golden? - The
We plan to support thousands of clients at the same Effects of Silence Deletion on the CPU Load of an
time. Audio Conference”, IEEE Proceedings of the
International Conference on Multimedia Computing
7. References and Systems, May 1994, Boston.
[1] International Telecommunication Union, “Packet [14] J. Junqua, B. Mak and B. Reaves, "A Robust
based multimedia communication systems”, Algorithm for Word Boundary Detection in the
Recommendation H.323, Geneva, Switzerland, Feb. Presence of Noise." IEEE Transactions on Speech
1998. and Audio Processing, vol.2, No.3, July 1994
[2] J. Rosenberg et al., “SIP: Session Initiation [15] Gerischer F., “Design and Implementation of Audio
Protocol”, RFC 3261, Internet Engineering Task Components for a Corba-based Multimedia
Force, June 2002, Platform”, Diploma Thesis, 1997, France.
http://www.ietf.org/rfc/rfc3261.txt. [16] International Telecommunication Union,
[3] The Access Grid Project, “Recommendation G.723.1 Silence Compression
http://www.accessgrid.org. Scheme”, Annex A, 1996.
[4] K. Singh and H. Schulzrinne. Interworking between [17] Elan Amir, Steven McCanne, Hui Zhang, An
SIP/SDP and H.323. In Proceedings of the 1st IP- Application Level Video Gateway, In Proceedings
Telephony Workshop (IPtel 2000), Berlin, of ACM Multimedia ’95 (Nov. 1995), ACM.
Germany, Apr. 2000. [18] A. Mankin, L. Gharai, R. Riley, M. Perez Maher,
[5] Jiann-Min Ho, Jia-Cheng Hu, Peter Steenkiste, A and J. Flidr, "The design of a digital amphitheater,"
conference gateway supporting interoperability In Proceedings of The 10th International Workshop
between SIP and H.323, Proceedings of the ninth on Network and Operating System Support for
ACM international conference on Multimedia, 2001 Digital Audio and Video (NOSSDAV 2000),
, Ottawa, Canada Chapel Hill, NC, June 2000.
[6] Virtual Rooms Video Conferencing System, [19] RTP: A Transport Protocol for Real-Time
www.vrvs.org Applications (IETF RFC 1889)
[7] Geoffrey Fox, Wenjun Wu, Ahmet Uyar, and Hasan http://www.ietf.org/rfc/rfc1889.txt.
Bulut, A Web Services Framework for
Collaboration and Audio/Videoconferencing, The

Videoconferencing in heterogeneous environments

More Related Content

What's hot (13)

Viewers also liked (6)

Similar to Videoconferencing in heterogeneous environments (20)

More from Videoguy (20)

Videoconferencing in heterogeneous environments