Communication Networks for Multimedia

1
Communication Networksfor Multimedia
2
The Evolution of Communication
Networks For over 100 years, the POTS (Plain
Old Telephone System) has been the primary
focus of conventional voice-band
communications POTS network is well designed
and well engineered for the transmission and
switching of 3-Khz voice calls Real-time
Low-latency High-reliability
Moderate-fidelity
3
Packet Networks POTS network is not
designed for other forms of communications,
such as wide-band speech, audio, images, video,
facsimile and data. About 30 years ago, a
second communications network was created with
the goal of providing a better transport
mechanism for data networking. The resulting
network is called a packet network because data
is transmitted and routed along the network in
the form of units of information
4
Packet Networks (contd) Packet networks
evolved independently of telephone networks for
the purpose of moving bursty, non-realtime data
among computers. Packets consist of a header
(information about the source and destination
addresses) and a payload (actual data being
transmitted). Packet networks are especially
well-suited for sending data of various types,
including messages, facsimile, and still
images. Packet networks are not well suited for
sending real-time communication signals such as
speech, audio and video.
5
The Open Systems Interconnect (OSI)
Architecture Physical layer is concerned with
the transmission and reception of unstructured
bit stream over any physical medium. It deals
with the mechanical aspects and signal voltage
levels. Examples are RS-232-C and X.21
Datalink layer ensures reliable transfer of data
across the physical medium. It also provides
access control to the media in the case of
local area networks. Examples are High-level
Data Link Control (HDLC), LLC and SDLC Network
layer provides the upper layers with
independence from the switching technology. It
is responsible for establishing, maintaining
and terminating connections. It is also
responsible for routing. Examples are X.25 and
IP
6
The OSI Architecture (contd) Transport
layer is responsible for reliable and
transparent transfer of data between end
points, takes care of end-to-end flow control
and end-to-end error recovery. An example is
TCP Session layer provides a means for
establishing, managing and terminating
connections between processes. It may also
provide checkpoints, synchronization and restart
of service Presentation layer performs a
transformation on the data to provide a
standardized interface to applications. It help
to resolve the syntactic differences when the
internal representation of the data differs
from machine to machine Application layer
provides services that can be used by user
applications.
7
Media Access Control Media Access
Control (MAC) systems may be divided into
three categories Round robin each
station on the network, in turn, is given
an opportunity to transmit. When it is
finished it must relinquish its turn and the
right to transmit passes to the next
station in logical sequence. Control of
turns may be centralized or distributed.
Token ring is an example of such scheme
MAC The lower sublayer of the OSI data link
layer. The interface between a node's Logical
Link Control and the network's physical layer.
8
Media Access Control (contd) Reservation.
Typically, the time on the medium is divided
into slots (time-division multiplexing). To
transmit, a station reserves future slots for an
extended or indefinite period. Shared satellite
channel is an example of this scheme
Contention. No control is exercised to determine
whose turn it is to transmit. These methods are
likely to lead to collisions and may require
retransmission. These techniques perform well
when the network load is less, progressively
drop off at moderate loads, and perform poorly
at high loads. CSMA/CD is an example
9
Network and Transport Layers Together, the
network and transport layers establish a data
pipe between the source computer and a process
on the destination computer the network
layer is responsible for setting up routes
from a source node to a destination node
the transport layer handles end-to-end issues
between processes running on this nodes, such
as error control, sequencing, flow control
10
Unicast and Multicast Multimedia
communications involve two basic modes unicast
and multicast In unicast mode there are two
communication partners, or peers, and the
resulting mode is called peer-to-peer
communications Example individual
client-server applications (home shopping,
video on demand) Multicast mode involves 1 to
N communications (peer-to-multipeer) as well as
1 to all communications (broadcast mode)
Examples distance learning, multipeer
teleconferences
11
Routing Network graph of nodes
(subnetworks) and edges (links between
subnetworks) Problem find an optimal path
from a given source to a given destination
node Routing is the problem of the main task
of the network layer and involves two major
subproblems Find an optimal path in the
routing graph, under changing network loads
and perhaps even a changing network
topology Get all incoming packets through a
router at runtime in an optimal way
12
Approaches to Routing Connectionless the
pathfinding algorithm is executed every time a
packet is injected into the network (e.g., IP).
Each packet finds its way independent of other
packets and carries a destination address.
Efficient for short connections (no connect
and disconnect phases in the protocol)
Robust in the case of a node failure (no state
information stays in the nodes) Easy
internetworking
13
Approaches to Routing (contd)
Connection-oriented (aka virtual circuit) a
path from source to destination is computed
only once for the duration of a connection. All
packets of a connection follow the same path
through the graph (e.g., X.25. Frame relay,
ATM) The connection set-up packet leaves
a trace with routing information in
each node on the path (a connection
identifier plus an output port), and all
subsequent packets follow the same path.
Efficient routing at runtime (no pathfinding
algorithms to be executed during a
connection) The ability to use access
control to avoid network congestion (a
new call is rejected when the network is
overloaded)
14
Routing Algorithms Static routing
all routes are pre-computed for a given
topology and are independent of the current
network load Each node has a table with
entries in the form source destination
outgoing link An incoming packet contains
the destination address (or, in the case
of virtual circuits, the connection
identifier) Routing decision is reduced to
a quick table look-up When the network
topology changes, a network control center
re-computes the global routing table, and
the new table is downloaded into all nodes
15
Routing Algorithms (contd) Adaptive
routing the path-finding algorithm
automatically takes into account new or obsolete
nodes and links as well as the current load of
nodes and links. Each node gets some limited
information from neighboring nodes and/or
extracts information from packets underway.
16
Broadcast Routing Send a distinct packet
to each destination Bandwidth wasteful
Requires the source to have a complete list of
all destinations Flooding Every
incoming packet in a node of the subnet is
sent out to every outgoing line except the one
it arrived on Must have a way to
dump the number of duplicate packets
E.g., each router can keep track of which
packets in a sequence have already been
sent
17
Broadcast Routing (contd)
Multi-destination Routing Each packet
contains a list of destinations New copies
of the packets are generated at each router
for the output lines that are needed After a
sufficient number of hops, each packet will
carry only one destination and can be
treated as a normal packet Spanning Tree
Broadcasting Spanning Tree (ST) subset of
nodes of the subnet that includes all the
routers but no loops If each router knows
which of its lines belong to the spanning
tree, it can copy an incoming broadcast packet
onto the ST lines except the one it arrived
from Most efficient but all routers must
know the ST
18
CBR and VBR Traffic Multimedia traffic
can be characterized as constant bit rate (CBR)
or variable bit rate (VBR) For CBR
applications, it is important that the network
that transports the data streams has a constant
throughput (otherwise, extensive buffering
would be required at each end of the system)
VBR traffic often occurs in bursts or spurts
(typical case video compression) A good
measure of burstiness is given by the ratio
of peak traffic rate over mean traffic rate over
a given period of time
19
CBR and VBR Traffic (contd) Even with CBR
networks, the throughput may vary with time due
to the following reasons Node or link
failure Network congestion (when the
demand for network capacity exceeds the
availability) Throughput decreases with
increasing load, especially when
bottlenecks are present in the network
Flow control . It is an end-to-end protocol that
places limits on the rate of data
transmission between two end-systems
connected through a network in order to
prevent loss of data at the receiving
end-system due to buffer overflow
20
Congestion and Flow Control Congestion
happens when too many packets are present in (a
part of) a subnet (performance degrades)
Congestion control makes sure that the subnet
can carry the offered traffic. Flow control
makes sure that a fast sender cannot
continuously send data faster than the receiver
can absorb it (involves feedback from the
receiver)
21
Congestion and Flow Control (contd)
Examples Fiber optic network at 1000 Gb/s
on which a fast computer is trying to
transfer a file to a PC at 1Gb/s There
is no congestion but flow control is required
Network with 1 Mb/s lines and 1000 computers,
half of which are trying to transfer files
at 100 Kb/s to the other half
There are not fast senders overpowering slow
receivers but the total offered traffic exceeds
what the network can handle (congestion)
22
Congestion Control Traffic
Shaping Sender and network agree on average
rate and burstiness of data transmission
It is not so important for file transfer but
very important for real-time data
(audio/video) which do not tolerate
congestion well Needs traffic policing to
monitor the traffic flow and make sure that the
customer is following the agreement
23
Example Leaky Bucket The Leaky Bucket
algorithm consists of a token counter and a
timer. The counter is incremented by one at each
T units of time and can reach a maximum value
C. A packet is admitted into the network if
and only if the counter is positive. Each time
a packet is admitted, the counter is
decremented by one. The traffic generated by a
Leaky Bucket regulator consists of a burst of
up to C packets followed by a steady stream of
packets with minimum inter-packet time of T
Parameters Capacity C (packets or bytes)
Flow ? (packets or bytes per second)
24
Leaky Bucket (contd) Example A
computer can produce data at 25 MB/s. The
routers can handle this data rates only for
short intervals. For longer intervals, they
work best at less than 2MB/s. Data
comes in 1MB bursts (one 40 ms burst every
second) To reduce the average rate to 2MB/s,
we could use a leaky bucket with ?2MB/s
and capacity C1MB This means that
bursts up to 1MB can be handled
without data loss, and that such bursts are
spread out over 500 ms, no matter how
fast they come in.
25
Flow Control Flow control is
typically performed using the Sliding Window
mechanism The sliding-window algorithm
allows the sender to transmit packets at
its own speed until a window of size W is
used up. It then has to stop and wait until
acknowledgments from the receiver open the
window again. In the TCP protocol, W is
not counted in terms of packet but in
terms of bytes in transfer
26
Sliding Window
27
Requirements and Performance The three most
important parameters of a communication network
for multimedia communications are
Throughput Error rate Delay They
form the basic parameters of the Quality of
Service (QoS)
28
Throughput The throughput of a network
corresponds to its effective bandwidth or bit
rate, i.e., the physical link bit rate minus
the various overheads Example ATM
technology over a SONET (Synchronous
Optical NETwork) fiber optics
transmission system. The network carriers
provisioned bit rate is 155.52 Mb/s. Principal
overheads are approximately 3 for SONET
and 9.5 for ATM. Thus, the maximum
throughput of this network is actually
136 Mb/s Other factors that affect throughput
are network congestion, bottlenecks, node or
line faults
29
Error Rate Defined in terms
of the bit (packet) error rate, i.e., the ratio
of the average number of corrupted bits
(packets) to the total number of bit (packets)
transmitted Examples In fiber optics
transmission, the bit error rate range from
10-8 to 10-12 In satellite transmission
systems, the bit error rate is on the order
of 10-7
30
Causes of Errors in Packet-Switching
Systems Individual bits in packets are
inverted or lost Error-correction codes are
able to correct the error or detect it and
request retransmission Bit error recovery is
based on error detection and
retransmission The sender learns about bit
error in one of two ways The receiver sends a
negative acknowledgment (NACK)
The sender signals a time-out unless a positive
acknowledgment is received within a
predefined interval
31
Causes of Errors in Packet-Switching Systems
(contd) Packets are lost in transit
(inadvertent error), dropped by an intermediate
node (deliberate error) or delayed In a
connection-oriented network, when packets are
lost or dropped, the receiving end-system
is usually able to detect such a situation
and inform the sending side Packet loss
recovery is based on sequence numbers In the
case of connectionless networks, packet loss or
dropped packets are difficult, if not
impossible to detect The primary reason for
packets being dropped or lost in high-speed
networks is insufficient buffer space at the
receiving end-system due to congestion in the
network
32
Causes of Errors in Packet-Switching Systems
(contd) Packets arrive out-of-order It is
the job of the receiving end-system to
rearrange the received packets in the numerical
sequence in which they were originally sent
IMPORTANT packet retransmission (especially if
it has to be carried out on an end-to-end basis)
significantly increases latency For
real-time video or audio transmission, delay is
a more important performance issue than
error rate, so in many cases it is
preferable to forget the error and simply
work with the received data stream as is
33
Delay (Latency) End-to-end delay is formed
by Network delay, composed of
transit delay which depends on the physical
distance between the two ends
transmission delay which is the time required to
transmit a block of data and depends on
the bit rate and on processing delays
in the intermediate nodes, including
routing and buffering Interface delay,
which is the delay incurred between the
time a sender is ready to begin ending a
block of data and the time the network is ready
to transmit the data
34
Delay (Latency) (contd) For
connection-oriented networks, when end-to-end
acknowledgments are required, round-trip delay
is useful Round-trip delay is defined as the
total time required for a sender to send a
block of data through a network and receive an
acknowledgment that the block was received
correctly
35
Delay Variation (Jitter) Extremely important
for synchronous multimedia streams (e.g., audio
and video) Network traffic can be
Asynchronous (no upper bound to the latency)
Synchronous (an upper bound to the latency
exists) Isochronous (there is a constant
transmission delay for each message, i.e.,
if two data streams traverse the network at
essentially the same rate and arrive at the
destination at the same time) Isochrony may be
recovered by an appropriate playout buffer at
the destination
36
Quality of Service (QoS) QoS
indicates how well a network performs in
dealing with a multimedia application
Individual applications have different
expectations of how well the network carries
out its tasks Real-time conferencing may
impose QoS requirements on latency and
throughput Downloading a video might
require small error rates but not have
tight restrictions on latency or throughput
37
QoS (contd) ? Resource Reservation and
Scheduling ? If an application knows in
advance that it requires certain QoS
resources it can make a reservation with the
network for those resources for the
period in question. The network can
either deny the request or schedule the
application for that period and reserve
the resources requested ? Resource Negotiations
? If the network administrator feels that the
requested resources might overtax the
capabilities of the network, it can
negotiate with the requester and offer lower QoS
parameters. A mutually acceptable set of
QoS parameters can then be negotiated.
38
QoS (contd) Admission Control If the
QoS demands of the particular application are
so high that the network cannot meet them, the
network has the choice of not letting the
application on to the network. ?
Guaranteed QoS ? The user may expect a
guaranteed level of service from the
network. Whether these guarantees are
statistical or absolute depends upon the
negotiations between the user and the network.
39
Example of QoS Requirements for Audio
40
Example of QoS Requirements for Video
41
Media Filtering In a multicast scenario,
not all receivers have the same QoS
requirements E.g., a PC connected via a
telephone line will not be able to
receive video at the same rate as a
highend UNIX workstation connected via ATM A
solution media filtering The internal
network nodes implement media filters, so
that the sender needs to create only one flow
satisfying the maximum QoS, saving
considerable bandwidth
42
Media Filtering - Example
43
Media Scaling A problem with a static
QoS contract between the sender and receivers
of a multicast stream is the variance of many
parameters throughout the duration of the
transmission, both at the end notes and within
the network. It would be desirable to adjust
the QoS parameters during a multimedia
connection. When applied to a multimedia data
stream, this is called media scaling.
44
Media Scaling (contd) Media scaling
allows control of parameters other than just
the data rate (which was already supported by
traditional connection-oriented protocols via
flow control mechanisms). Example image
quality in video stream To implement media
scaling, the interface between the application
and the network must be extended to pass
control information. Example if the
network signals increasing congestion, an
MPEG video encoder can adjust its data
rate out via any of its scalability techniques.