Kari Melkko
Department of Computer Science and Engineering
Helsinki University of Technology
Kari.Melkko@hut.fi
Till late 80's Internet was mainly used for tasks that were text based services such as virtual terminal sessions or non-realtime tasks such as e-mail and file transfer. These kinds of applications are considered narrowband, which means that network connection with primary (comparatively low) data rate is needed for smooth usage.
In the beginning of 90's increasing number of Internet-connected hosts equipped with graphical user interfaces and elementary multimedia capabilities set new demands on net applications. Users wanted to integrate multimedia features and Internet connectivity. One of the first and best known multimedia applications were WWW-browsers which made extensive use of hypertext transfer protocol (HTTP) developed by Tim Berners-Lee at CERN year 1989. First graphically oriented WWW browsers supported only text and still images but demand for real multimedia lead to current situation where browsers are equipped with numerous add-on applications that handle realtime audio- and videostream. Unconnected to HTTP, several other applications including tools for videoconferencing and telephony services over Internet have become available.
Common factor to these new applications is that they deal with realtime streams and they usually require certain well-known network bandwith often exceeding 100Kbps to function properly. Such requirements imply that these applications are considered broadband. Internet protocols were not designed to handle traffic caused by broadband applications; the main design principle was to assure reliability and interoperability between network elements from different vendors.
This paper's aim is to clarify what kind of broadband applications are
under development or in use, their special characteristics and requirements
on Internet, and to consider some proposals and visions how Internet and
Internet protocols designed in 70's could meet these new demands.
A and B would be easy to solve with some buffering and error correction protocol that resends dropped packets, but because of C stream cannot be buffered (much) in either end. C also directs the max. packet size:
Tp + (packet size / link speed) + Tbuf = Td < 500ms
(Tp = packet propagation delay, Tbuf = buffer size in seconds, Td =
total delay)
Since transmission of the stream is (almost) unbuffered, variation in packet propagation delay Tp is unacceptable. Any jitter causes problems with sound quality; a packet arriving too late creates pause and a packet arriving too early is discarded. Sophisticated Internet telephony applications compress the audio and add some redundancy to packets resulting a stream that is not so critical, but however the requirements on the network are quite the same:
Videoconferencing requires synchronization between audio,video and additional data. One way to meet those requirements is to use a framing protocol that combines separate streams. An example of videoconferencing protocols based on framing is ITU-T's H.320 standard. Packets transmitted during a H.320 videoconference are formatted using a framing protocol that sends 80-byte packets across the network. The first few bytes of a videoconference call will look like[15]:
AADDVVVS
AADDVVVS
AADDVVVS
AADDVVVS
AADDVVVS
A = audio, D = data, V = video, S = service channel
This means that H.320 produces a Constant Bit Rate stream, and the requirements are quite similar to telephony applications. H.320 was originally designed for ISDN and PSTN environment having only narrowband CBR service. In broadband network it would be more efficient to send video in separate variable bit rate stream; rapid movements of participants generate transmission peaks but most of the time image is still and there is no sense to send data. However, having three separate streams, CBR audio, Bursty data channel and VBR video, that must be synchronous raises new demands on network. Network should provide for all these streams similar (minimal) transmission delay, but if congestion is about to happen in transmission path, network should first discard packets belonging to video stream, then data and last audio. In addition, videoconferencing is usually one-to-many or many-to-many communication so there is need for multicast possibility. Requirements:
| sound quality | bandwidth | mode | bitrate | reduction ratio |
|---|---|---|---|---|
| "telephone sound" | 2.5 kHz | mono | 8 kbps | 96:1 |
| "better than shortwave" | 4.5 kHz | mono | 16 kbps | 48:1 |
| "better than AM radio" | 7.5 kHz | mono | 32 kbps | 24:1 |
| "similar to FM radio" | 11 kHz | stereo | 56...64 kbps | 26...24:1 |
| "near-CD" | 15 kHz | stereo | 96 kbps | 16:1 |
| > "CD" | 15 kHz | stereo | 112..128kbps | 14..12:1 |
Usually there are no disadvantages if the audio stream is buffered. 1-2 seconds buffer solves many problems like lost packets, variation in transmission delay and temporary bandwidth insufficiency. Broadcast applications (net radio etc.) require multicast possibility which reduces network load efficiently. Requirements:
every MPEG-1 compatible decoder must be able to support at least video
source parameters up to TV size: including a minimum number of 720 pixels
per line, a minimum number of 576 lines per picture, a minimum frame rate
of 30 frames per second and a minimum bit rate of 1.86 Mbits/s.
MPEG-2 provides video quality not lower than NTSC/PAL and up to CCIR
601 quality. MPEG-2 implementations will at least conform to the MAIN Profile
at MAIN Level which supports non-scalable coding of digital video with
approximately digital TV parameters - a maximum sample density of 720 samples
per line and 576 lines per frame, a maximum frame rate of 30 frames per
second and a maximum bit rate of 15 Mbit/s.[MPEG]
Both algorithms produce variable bit rate (VBR) stream.
Because bandwidth of videostreams usually exceeds 2 Mbit/s, buffering may become a problem despite the fact that stream contains so much redundancy that a few dropped or erroneous packets don't usually lower the video quality significantly. Adaptation to temporary network bandwidth limitations and variation on transmission delay are the main reasons for buffering. Therefore network should provide variable bitrate service which means that enough bandwidth is reserved for some constant bit rate component and agreed amount of peaks which may not exceed some negotiated limit in time unit. Service should provide minimum possible variation on transmission delay. As with audio stream applications, multicast support is required to minimize network load. Requirements:
When user moves in VRML world loads of network traffic is generated; while he/she is quiescent no data is sent. To allow smooth traveling in virtual reality data should be sent with maximum bitrate and minimum delay. This requires the network support for high bit rate bursts with minimum possible delay. Because VRML documents reference objects that are distributed on many servers, connection setup phase may cause considerable portion of total transmission delay. In short, the requirements are:
Especially multicast support in IP level is feature that broadcast
applications, for example net radio and live video stream, require.
Congestion-controlled (C-C) traffic is not so critical, it is acceptable that there is variable amount of delay in delivery of packets and packets may arrive out of order. C-C traffic is usually backed up by some transport protocol, an example being TCP.
Non-congestion-controlled traffic is usually the traffic that realtime broadband applications handle. As seen in chapter 3, for this kind of traffic constant data rate and a constant delivery delay (or at least relatively smooth data rate and delay variation) are desirable. Usually it makes no sense to retransmit dropped packets. In addition, router should not hold N-C-C packets in buffers too long but discard some packets (in priority order) in case of bandwidth insufficiency.
Priority field in itself is a great improvement when considered the requirements of broadband applications on the Internet.
Also LAN access seems to have bandwidth problems. 10Mbps Ethernet bandwidth is shared by too many users. Currently, majority of large corporate Internet users are connected at below 128Kbps [2]. Ethernet switching provides assured 10Mbps to each user assuming that switch is connected to backbone that has enough capacity. Alternative way is to upgrade LAN to 100Mbps or 1Gbps. Drawback in that scheme is that Ethernet does not support QoS; Access to transfer media is controlled by CSMA/CD that may cause substantial variation on transfer speed.
Albeit ATM seems fine, it has severe weaknesses: ATM was designed mainly
by Telcos, for Telcos. There has been a lack of equipment suitable for
single user and if equipment is available it is considerable overpriced
when compared to products of other technogies. Also ATM and IP integration
has been painful. One may ask why ATM and IP have to be integrated and
do we really need IP at all anymore? Answer is obvious; Only few applications
have been developed for ATM. Increasingly broadband software is being built
to interface with the IP-based Internet, intranets and extranets. Besides,
the Internet continues to enjoy extremely high growth rates of almost every
associated measure (for example, users, traffic and connected networks)
whilst ATM is still suffering "the ATM vicious circle" [3].
There have been few proposals on ATM exploitation in IP context, the
most interesting ones beeing LANE and multi-protocol over ATM (MPOA). LAN
emulation allows LAN traffic to be carried transparently over an ATM network.
LANE allows an easy migration, but means that sophisticated ATM features
needed by broadband applications are not exploited. MPOA is more elaborate
solution, it allows layer 3 protocols (IP) to operate directly over ATM.
The MPOA solution maps routed and bridged flows of traffic to ATM switched
virtual channels (SVCs), off-loading traditional routers from performing
packet-by-packet processing. Furthermore, ATM's built in QoS benefits can
be realized for multimedia applications that involve continuous flows of
voice and video traffic that requires bandwidth guarantees. The result
is a multi-gigabit routing infrastructure that meets the demands associated
with some broadband applications. As ATM is connection oriented there is
noticeable initial delay before real data transmission begins. This initial
delay may be unacceptable for some applications.
Routers decide which packets are sent first, they choose paths and decide
which packets are discarded in case of congestion. Routers hold resources
that are needed to assure some quality of service. Applications should
be able to reserve resources (bandwidth, buffers, processing time, discard
policy) from the routers on transmission path. To enable this, some resource
reservation protocol has to be widely adapted in Internet infrastructure,
for example RSVP. Routers should support flows with additional QoS parameters
and they should assure some short initial delay for flow setup.
When increasing Internet capacity routers play significant role because
at present they restrict bandwidth. Currently average packet size in Internet
is about 2000 bits [5]. To exceed 1Gbps data rate routers have to forward
over 500 000 packets/s. This leaves less than 2µs for packet processing.
One way to achieve faster routing is IP switching tecnique, which combines
ATM hardware and IP router functions enabling some interesting features
for broadband applications needs.
A scheme based on IP switching routers would enable classic, narrowband
Internet and new, broadband Internet co-exist.
BER, Bit Error Ratio
BPS, Bits Per Second
CBR, Constant Bit Rate
CCIR, International Radio Consultative Committee, standard 601
describes a digital coding standard for television that is applicable to
both the NTSC as well as PAL/SECAM technologies
ISDN, Integrated Services Digital Network
ITU-T, International Telecommunication Union, Telecommunications
sector
NTSC, Colour television system used in USA and Japan. 525 lines,
60Hz frame rate.
PAL, Colour television system used in Northern and Western Europe.
625 lines, 50Hz frame rate.
PSTN, Public Switched Telephone Network
QOS, Quality of Service
TCP, Transfer Control Protocol. Connection-oriented transfer
protocol of TCP/IP Protocol Suite.
UDP, User Datagram Protocol. Connectionless transfer protocol
of TCP/IP Protocol Suite
VBR, Variable Bit Rate
[2] Stallings, W. "High Speed Networks, TCP/IP and ATM design principles",
Prentice-Hall Inc., 1998.
[3] Matthews, J. "The Future of Broadband Networking, ATM versus TCP/IP",
Ovum Ltd., 1997.
[4] [RFC-1883] Deering, S., and Hinden, R., "Internet Protocol, Version
6, Specification", RFC 1883, Xerox PARC, Ipsilon Networks Inc., December
1995, p. 37.
[5] Newman P., Minshall G., Lyon T, and Huston L., Ipsilon Networks
Inc. "IP Switching and Gigabit Routers", IEEE Communications magazine January
1997.
[6] Extreme Networks, Multi-gigabit routers, <http://www.extremenetworks.com/>.
[7] Nokia Telecommunications, IP Switching, <http://www.ipsilon.com/>.
[9] ATM Forum, ATM related articles, <http://www.atmforum.com/>.
[10] The ADSL Forum Office, ADSL modems, <http://www.adsl.com/>.
[11] MPEGTV, MPEG compression, <http://www.mpeg.org/>.
[12] Helsinki Telephone Company, Interactive Virtual Reality Helsinki,
<http://www.helsinkiarena2000.fi/>.
[13] VRML Consortium, VRML specifications and standards, <http://www.vrml.org/>.
[14] Intel Corporation, H.324 Video Phone Standard, <http://www.intel.com/pcoems/psvideo/h324whit.htm>.
[15] Roberts, M., H.320 Video Conferencing Protocol, <http://bugs.wpi.edu:8080/EE535/hwk97/hwk4cd97/bigles/bigles.html>.
When compared to the requirements that broadband applications raise,
ATM seems to be the right choice. It delivers a connection oriented service
with assured quality, while exploiting packet efficiences.
The
ATM vicious circle.
The Internet virtuous circle.
6. Need for better connections - switching or routing?
First of all, to allow broadband application use in large scale, Internet
capacity should be increased considerably. There is also a strong need
to be able to support a variety of traffic with a variety of QoS requirements
and to implement multicasting, within the TCP/IP. The key elements of the
TCP/IP networks when implementing these new features are routers.
6.1 multi-gigabit routers
Multi-gigabit routers are based on forwarding engine, switch fabric and
network processor that runs routing protocols and updates routing tables.
Forwarding engine performs routing table lookup, rewrites the processed
IP packet and passes it to switch fabric which delivers packet to outgoing
line. Switch fabric is the component enabling speeds in excess of 1Gbps.
Gigabit routers will provide sufficient speed so that ATM switching will
not be required; TCP/IP can operate directly on SDH/Sonet using routers
with optical interfaces. Gigabit routers provide sufficient speed for broadband
applications, but at present they lack QoS and multicast features. QoS
processing also slows down forwarding engines, resulting lower capacity.
Same applies to multicasting. Gigabit routers don't support flow concept,
they treat every packet individually.
6.2 switched IP
IP switching is an alternative to the gigabit router. An IP switch maps
the forwarding functions onto a hardware switch, often an ATM switch. Switched
IP approach uses the concept of a flow, which enables QoS features to be
fully supported. IP switch maps incoming flows to ATM virtual channels
established across the ATM switch fabric. Packets are forwarded using classical
forwarding engine until ATM VC is established. This procedure hides long
ATM initial delay. Once the virtual channel is established, all further
traffic on that flow can be switched directly through the ATM switch. This
reduces load on the router and enables very high speed. Switched IP fully
utilizes sophisticated ATM features such as multicast and QoS. Even better,
connected hosts don't have to know anything about ATM. Switched IP appears
to be very promising technology that enables network infrastructure needed
by broadband applications.
7. Conclusion
Altough the requirements of broadband applications call for numerous improvements
and changes to existing Internet infrastructure, it seems that the transition
from narrowband Internet to broadband is going to happen soon. At present,
technology to implement the new features and speeds exists, IPv6 and RSVP
protocols that enable full utilization of underlying sophisticated hardware
are defined and some broadband applications are already in use. Main reason
for quick realization of broadband Internet is that better infrastructure
enabling widespread use of broadband applications and resource reservation
with provided QoS opens new business opportunities.
8. Glossary
ATM, Asynchoronous Transfer Mode
references:
[1] Huitema, C. "IPv6 The New Internet Protocol", Prentice-Hall Inc., 1996.
