Quality of service in multimedia computer networks.
Macura, Arijana ; Missoni, Eduard ; Makovic, Branko 等
Abstract: Implementing a specific algorithm for synchronization
based on a selection of relevant QoS (Quality of Service) parameters
requests their detailed analysis in order to enable realization of the
desired multimedia applications. Ensuring of requested QoS presents one
of the important elements of the design and implementation of multimedia
networks. The paper analyzes the parameters of importance for multimedia
communications. In recent years there has been a rapid increase in
multimedia traffic in computer communication networks and the demand for
quality of multimedia content transfer services for the users. In
response to such requests numerous mechanisms have been developed that
enable services for allocating priority of traffic in communication
networks, which makes it possible to determine the quality of service.
Key words: algorithm, TCP/IP, reno, tahoe, SACK
I. INTRODUCTION
Implementing a specific algorithm for synchronization based on a
selection of relevant QoS (Quality of Service) parameters requests their
detailed analysis in order to enable realization of the desired
multimedia applications. Ensuring of requested QoS presents one of the
important elements of design and implementation of multimedia networks.
Quality of service is the ability of network to provide better service
to certain types of network traffic. This ability can be achieved with
different network technologies, as well as in the layers of different
network protocols (usually TCP/IP protocols). TCP protocols makes
possible a reliable transfer of connection type. Primary TCP would
initiate connection by manifold transmitter segments pricking to the
network up to the window size notified by the receiver. That is
acceptable when two hosts are in the same LAN, but in the situations
when there are routers and slower links between the transmitter and
receiver problems can emerge. Particular mediating routers must create
waiting files with packages in such a way that mentioned router remains
without free space for the temporary location of the package. In order
to avoid that problem, it is necessary to develop slow-start algorithm.
Congestion can happen when manifold incoming flows arrive on router
whose outgoing capacity is smaller than the sum of all entrances. For
that reason algorithms are developed to avoid congestion. Quality of
service in computer network traffic is the ability of communication and
transmission devices to guarantee the minimum requirements, in other
words, quality of traffic required by some network application. TCP is
based on principle of package "preservation", which means if
connection works on available capacity of the permission size, in that
case package is not thrown in net before the other package leaves the
net. As distributed multimedia systems grow larger and faster, the
associated problems of resource management become harder to manage. The
traditional method of managing applications is that, prior to execution,
an application determines its quality of service (QoS) requirements in
terms of various parameters such as image and sound quality.
2. FUNDAMENTALS OF IMPROVING MULTIMEDIA DATA TRANSFER IN TCP/IP
COMPUTER NETWORKS
Frequently used or suggested improvement solutions for multimedia
data transmission quality over TCP/IP computer networks are divided into
four main categories.
2.1 Solving big bandwidth-delay product
Most solutions in this area have emerged because of the necessity
to solve the problems of older TCP/IP implementations. In the older
implementations there is a problem of solving big bandwidth-delay
product because of a small congestion window parameter, a slow sliding
window algorithm and also slow recovery time from traffic congestion.
Also, the overall scalability of the transfer depending on the number of
simultaneous data flows is one of the most important areas for improving
the TCP/IP algorithm. Some proposed changes include increased initial
TCP window size, changes in sliding window algorithm, and the
introduction of selective acknowledgment (SACK) that allows proper
adaptation of the TCP source of network traffic in case of loss of more
network frames within one RTT ( Round Trip Time) period ( Fall &
Floyd, 1996). Given that the increase of "congestion window"
parameter is based on increase of the number of frames and not on frame
length (in Bytes), therefore it is recommended to know beforehand the
maximum frame size on any given segment or connection. Usage of the
maximum permitted frame size affects the delay, but it also maximizes
the usage of network bandwidth by reducing the impact of header -
overhead. TCP flow and congestion control algorithms present the problem
for the transmission of multimedia data due to the basic problem that
the connection constantly increases network load until it reaches the
level of congestion, after which the load rapidly decreases, and then
the process repeats in iteration. Load (bandwidth) variation and the
network congestion example can be shown with case of two flows of data
that use the same portable algorithm (e.g. TCP Reno), with the same
network delay and transmission initiation's beginning. Also, given
that in this situation the total of network's bandwidth is is used,
and the data flow with the bigger delay does not have enough resources
for normal traffic, both TCP Reno flows will continue to compete for the
available bandwidth, steadily bringing the network into a state of
congestion (Casseti & Meo, 2001). This behavior of TCP Reno
implementation shows considerable intolerance toward connections with
bigger delays in the network (Gao & Zheng, 2001). Figure 1. shows
the comparison of the scalability of the main implementation of the TCP
protocol. The effect of such behavior on multimedia communications can
be extremely negative, resulting in a very low quality of video and
audio streams in a network that has sufficient capacity to support them
( in terms of bandwidth, delay, or variation in delay). Therefore it is
necessary to find a better algorithm for detection and control of
congestion inside protocol. Given that the framework for the protocol
allows the introduction of different algorithms for conduct in a state
of congestion, it is possible to increase efficiency and intelligence of
TCP by implementing new algorithms with adaptive behaviors. TCP Tahoe
algorithm for avoiding congestion utilizes "additional increase and
manifold decrease" (AIMD). Loss of package is considered as a sign
of congestion and Tahoe records half of temporary window as the limit
value, and then starts slow- start till the limit value is reached.
After that, increments line-ray till the loss of package happens. In
such a way the window size slowly increases till it comes nearer to the
full size capacity. Disadvantage of Tahoe algorithm is the need of
complete interval of time limitation to detect loss of a package. TCP
Reno is keeping back the basic principles of Tahoe algorithm, such as
slow- start, but detects earlier lost packages and flow structure is
discharged each time a package is lost. Reno requires getting
confirmation as soon as segment comes to destination. Reno suggests
algorithm named fast-retransmit which decreases transmitter waiting time
before lost segment retransmission. Reno behaves well when there are
small packages loses. But, when we have manifold packages loses in the
same window, its performances are approximately the same at those of TCP
algorithm at the same conditions. TCP SACK with "selective
confirmations" is widening in relation to TCP Reno and works on
problems which TCP Reno meets, especially on detection of manifold lost
packages, as well as retransmission of more than one lost package
through RTT. SACK keeps slow- start and fast- retransmit parts of TCP
Reno. It has the same long lasting time limitations as Tahoe, in case
that the package loss has not been detected by the modified algorithm.
TCP SACK required segments not to be confirmed cumulatively, but
selectively. The biggest problem with SACK is that momentary selective
confirmations are not supported by transmitter. SACK implementation or
better selective confirmations is not an easy task, but based on
simulation results can be concluded that TCP SACK shows the best
results. No one of the proposed versions realize the best results in all
circumstances. (Ceco & all., 2010). SACK behaves like TCP Reno when
the transfer takes place normally, or when there is a loss of one
network frame within one transfer window (Floyd & all., 2000).
However, when it comes to losing more than one network frame, SACK
allows faster recovery and greater achieved bandwidth. SACK achieves
better performance in all cases of transfer, with a score of over 95%
bandwidth utilization with a small amount of losses. Also, in case of
losing more than one network frame within one transfer window, SACK
implementation shows a faster recovery and better bandwidth than Reno
implementation. TCP SACK algorithm made minimal changes to TCP Reno
implementation. The key improvement is achieved by allowing selective
acknowledgement of received frames, and variable setting, by which it
tracks the remaining number of frames in transit through the network. In
this way allowing the algorithm to be undisturbed even more frames are
lost in the network within one transfer window. Dependence on measuring
network frame's round trip time (RTT) as the main parameter for
increase of data flow bandwidth in Reno and SACK implementations is
responsible for the preference of data flows with smaller delay, in
other words, RTT time.
[FIGURE 1 OMITTED]
TCP Tahoe algorithm's specific improvements are related to the
mechanism of detection and estimated network frame's roundtrip time
through the network Also, fast-retransmit algorithm built-in the Tahoe
implementation was later modified in modern implementations. If you
disable this function, TCP Tahoe transfer algorithm will take
considerable time to detect network frame loss and then send it back to
the network, which means that there will be decrease in transmission
quality if only one network frame gets lost. When frame loss gets
detected, transmission will continue with reactivation of slow-start
algorithm, which means that there will be a gradual increase in
transmission bandwidth.
3. CONCLUSION
Troubleshooting QoS management in multimedia computer networks is
based on discovering the efficient allocation of network resources, with
realization of QoS requirements and a high degree of resource
utilization. Different applications have different requirements for
their data transmission means. Applications will generally expect that
the network can transmit data at the speed at which they are generated.
In addition, applications show various sensitivity to many other
transmission parameters, such as delay and delay variation in the
transmission. Also, some applications are more sensitive than others to
the problem of packet loss, in other words, data fragments in the
transmission. These network applications' characteristics are
expressed through the parameters of capacity, delay, jitter and loss
rate. Priority and service quality allocation mechanism must be ensured
to meet the communications requirements of some applications, to reserve
the free capacity of computer network. The future researches will
continue in direction to discoveries of new mechanisms and creation of
new version of TCP protocol aiming in combination of existing versions
of TCP protocols in joint implementation, which would have superior
performances.
4. REFERENCES
Casetti, C. & Meo, M. (2001). Modeling the Stationery Behavior
of TCP Reno Connections, Proceedings of the International Workshop on
Quality of Service in Multiservice IP Networks, ISBN:3-540-41512-2, pp.
141-156, January 2001.
Ceco, A.; Bradic, K.& Nosovic, N. (2010). Performance
comparison of different TCP versions, Available from:
http://www.telfor.rs/files/radovi/TELFOR 2010-10-23.pdf Accessed:
2011/03/07
Fall, K. & Floyd, S. (1996.) Simulation--based Comparisons of
Tahoe, Reno and SACK TCP. ACM SIGCOMM Computer Communication Review,
Volume 26, Issue 3, July 1996., ISSN:0146-4833, ACM New York, USA
Floyd, S.; Maldavi, J.; Mathis, M. & Podolsky, M. (2000). An
Extension to the Selective Acknowledgement (SACK) Option for TCP,
Available from: http://www.icir.org/floyd/ talks/dsack-nov99.pdf
Accesed: 2011/03/11
Gao, C.; Li, M. & Zheng, F. (2001). An Analytic Throughput
Model for TCP Reno, Proceedings of the 2001 International Conference on
Computer Networks and Mobile Computing, ISBN 0-7695-1381-6, pp. 111,
October 16-19, 2001