P1: 211
Zarki WL040/Bidgolio-Vol I WL040-Sample.cls September 15, 2003 8:55 Char Count= 0
CURRENTSTATE OF THEINTERNET 835BGP is connection oriented and uses TCP as the transport
protocol. BGP supports IP prefixes and path aggregation,
which makes it suitable for CIDR.Real-Time Traffic Support
An important class of applications used over the Internet
is real-time. Real-time applications generate traffic that
must be communicated to the recipient(s) within a very
short amount of time. The time limit may be milliseconds
for interactive applications such as voice and video tele-
phony, but it may be larger for certain streaming or trans-
action processing applications. Despite increasing inter-
est in supporting such applications over the Internet, the
Internet is still largely a best-effort network without any
delivery guarantees necessary to provide the levels of qual-
ity of service (QoS) expected from end-users.QoS Support
Different applications typically have different traffic char-
acteristics and QoS requirements. Applications such as
streaming video usually require large bandwidth, and in-
teractive applications such as voice over IP (VoIP) and
video over IP (VIP) require tight delay bounds as the data
must be played back continuously at the rate they are
sampled. If the data (packet) does not arrive in time, the
playback process at the receiver will be disturbed. For ex-
ample, in voice telephony, human beings can tolerate a
latency of up to about 200 ms (Brady, 1971), although in
most of today’s voice networks, the latency is limited to
around 50 ms. If the latency exceeds this value, the degra-
dation in call quality will be noticeable. If enough extra
bandwidth is available, best-effort service may be able
to fulfill the delay, throughput, and other requirements.
When resources are scarce, however, real-time traffic will
suffer from congestion and delay, resulting in degradation
in the application quality.DiffServ.To facilitate end-to-end QoS on IP networks,
the IETF has defined two models: Integrated Services
(IntServ) (Braden, Clark, & Shenker, 1994) and Differen-
tiated Services (DiffServ) (Blake et al., 1998). IntServ was
defined first and follows the signaled-QoS model, where
the end-hosts signal their QoS needs to the network us-
ing resource reservation protocol (RSVP) (Zhang, Deer-
ing, Estrin, Shenker, & Zappala, 1993). RSVP signaling
and reservation of the desired QoS is done for each flow
in the network. A flow or a stream is defined as an individ-
ual, unidirectional data stream between two applications,
and is uniquely identified by a 5-tuple (source IP address,
source port number, destination IP address, destination
port number, and the transport protocol). While IntServ
provides for a rich end-to-end QoS solution, there are sev-
eral problems with the approach. State information for
each reservation needs to be maintained at every router
along the path, and each packet must be processed to en-
sure that the QoS of each flow is being satisfied. As there
may be hundreds of thousands of simultaneous flows go-
ing though a network core (router), it is not clear whether
IntServ will scale well in terms of complexity of admission
control, memory requirements of maintaining state, and
packet processing overhead.Since per-flow QoS is difficult to achieve in an end-
to-end fashion without introducing scalability issues, it
naturally leads one to think about classifying flows into ag-
gregates (classes), and providing QoS to aggregates rather
than to individual flows. For example, all real-time flows
could be grouped into a single class, and bandwidth and
other resources can be allocated for the class. This would
reduce the router’s burden on classification of traffic, sig-
naling, and state maintenance requirements. This is the
approach taken in the DiffServ model. In this model, pack-
ets are first divided into classes by marking the type of
service (ToS) byte in the IP header. A 6-bit bit-pattern
called the Differentiated Services Code Point (DSCP) in
the IPv4 ToS Octet or the IPv6 Traffic Class Octet is used
to this end. Once packets are classified at the edge of the
network, specific forwarding treatments, formally called
per-hop behavior (PHB), are applied at each network el-
ement, providing the packet with appropriate guarantees
(deterministic or statistical) on performance metrics such
as delay, delay jitter, and bandwidth. This combination of
packet marking and utilization of PHBs results in a more
scalable QoS solution.MPLS.Multiprotocol label switching (MPLS) (Rosen,
Viswanathan, & Callon, 2001) is another emerging tech-
nology that seeks to introduce QoS guarantees on the
Internet. Existing link state protocols, specifically OSPF
and IS-IS, provide the link state information about the
underlying IP network. Such information is used to de-
termine the best path through the network called label
switched paths (LSPs) using criteria such as number of
hops and other configurable parameters such as delay and
bandwidth.
An incoming packet to an MPLS network is assigned
a “label” by an “edge-label switch router” (Edge-LSR).
This label is swapped by intermediate label switch routers
(LSRs) as the packet traverses the MPLS network on an
LSP, and the final label is removed when leaving the MPLS
network. The label distribution protocol (LDP) is used
to establish label-to-destination network mappings. For-
warding of a packet is based solely on the contents of the
label, and not on the IP headers as is done in normal IP
routing, speeding up the process. Such increase in perfor-
mance, as well as the ability to perform traffic engineer-
ing, makes MPLS a strong contender for the converged
network.Real-time Transport Protocol (RTP)
The real-time transport protocol (RTP) (Schulzrinne, Cas-
ner, Frederick, & Jacobson, 1996) is an IP-based proto-
col providing support for the transport of real-time data
traffic such as video and audio streams. The services pro-
vided by RTP include timestamping, sequence number-
ing, and other mechanisms to take care of the timing is-
sues. RTP also provides information about the encoding
method used in the underlying data. Through these mech-
anisms, RTP provides end-to-end transport for real-time
data over the IP network. RTP was primarily designed for
multicast of real-time data; unicast is supported as well.
It also can be used for one-way transport such as video-
on-demand service as well as interactive services such as
VoIP and VIP.