17-03-2014, 07:12 PM
Abstract-Scheduling algorithms have significant role in
providing of the quality of service (QoS) in data networks.
Mostly, service rate is considered as an isolating parameter in the
scheduling algorithms. However, when a delay-sensitive and
bursty session such as video streaming or interactive gaming is
scheduled by these rate-based servers, the imposed delay may be
much more than an acceptable threshold which is required for
QoS provisioning. In this paper, we propose a fluid flow
scheduling algorithm which applies burstiness as an additional
isolating parameter. We assume that the arrival burstiness
constraint is leaky bucket. Our proposed scheduling algorithm
attempts to provide a service discipline similar to the arrival
constraint. In our algorithm, the weight of each session and also
service rate may increase when a burst arrives. Therefore, the
scheduler can isolate some bursty sessions to receive much more
amount of service in burst condition. The evaluation of the
proposed algorithm is carried out by calculating packet delay
statistics through a simulation strategy where various kinds of
traffic are scheduled by the proposed algorithm.
Index Terms-Scheduling, Quality of Service, Fairness, Bursty
Traffic, GPS.
I. [NTRODUCTION
Multimedia applications such as [PTV, gaming, Voice
Over [P (VO[P) and Video On Demand (VOD) are going to be
tremendously widespread among Internet and Intranet users.
These applications can provide a wide range of services
simultaneously. Some applications such as VO[P, generate a
constant bit rate (CBR) flow of traffic while others such as
compressed video transmission and web browser produce a
stream of traffic with variable bit rate (VBR). In addition, realtime
applications (e.g. VoIP and IPTV) are delay-sensitive
while other non-real time services (e.g. FTP and Email) do not
need any guaranteed upper bound of the end-to-end delay.
Therefore, service providing for real-time variable bit rate (rtVBR)
sessions which require a bounded delay is more
challenging in comparison with other applications.
Different applications have different traffic natures which
imply different QoS requirements. QoS is usually quantified by
a single or combination of measurable parameters such as end-
978-1-4799-0801-1/12/$31.00 ©2012 IEEE
to-end delay, bandwidth, delay jitter (variation in delays) and
packet loss. To provide guaranteed QoS in a network, not only
the cooperation among different layers inside a network device
is required, but also these appliances should collaborate with
each other in the network in order to appropriately assign their
available resources to the demanding sessions . .In network
infrastructure, routers and data switches are the most important
devices in QoS provisioning. One of the components of a
switch/router that plays a significant role in providing QoS is
the scheduling module. A scheduling algorithm is a method
that chooses the most appropriate packet among competing
packets for transferring to a specified output port. A QoS
scheduling algorithm is the one that considers QoS parameters
in its decision making process. Thus, coincident with the
emergence of multimedia applications in recent years, many
researches have been conducted to achieve high performance
QoS scheduling algorithms on multimedia applications [1]-[6].
A well-known class of scheduling algorithms is Rate
Proportional Server (RPS) which isolates different sessions
from each other based on their service rates [7]. An ideal
scheduling algorithm in this class is Generalized Processor
Sharing (OPS) [8] which is used as a benchmark in evaluation
of practical algorithms. Some other types of packet-by-packet
scheduling disciplines such as SCFQ [9] and FFQ [10], try to
approximate OPS in real life.
When all the incoming traffic to the scheduler is constant
bit rate, an RPS server, say OPS, ideally share bandwidth
among them. But when an RPS server needs to schedule a
mixture of VBR and CBR flows, the provided QoS for rt-VBR
(bursty and delay-sensitive) sessions may be unacceptable. The
reason is that the packet delay in a VBR flow may be much
more than the packet delay in a CBR flow. We show this fact
by an example in section [I.
providing of the quality of service (QoS) in data networks.
Mostly, service rate is considered as an isolating parameter in the
scheduling algorithms. However, when a delay-sensitive and
bursty session such as video streaming or interactive gaming is
scheduled by these rate-based servers, the imposed delay may be
much more than an acceptable threshold which is required for
QoS provisioning. In this paper, we propose a fluid flow
scheduling algorithm which applies burstiness as an additional
isolating parameter. We assume that the arrival burstiness
constraint is leaky bucket. Our proposed scheduling algorithm
attempts to provide a service discipline similar to the arrival
constraint. In our algorithm, the weight of each session and also
service rate may increase when a burst arrives. Therefore, the
scheduler can isolate some bursty sessions to receive much more
amount of service in burst condition. The evaluation of the
proposed algorithm is carried out by calculating packet delay
statistics through a simulation strategy where various kinds of
traffic are scheduled by the proposed algorithm.
Index Terms-Scheduling, Quality of Service, Fairness, Bursty
Traffic, GPS.
I. [NTRODUCTION
Multimedia applications such as [PTV, gaming, Voice
Over [P (VO[P) and Video On Demand (VOD) are going to be
tremendously widespread among Internet and Intranet users.
These applications can provide a wide range of services
simultaneously. Some applications such as VO[P, generate a
constant bit rate (CBR) flow of traffic while others such as
compressed video transmission and web browser produce a
stream of traffic with variable bit rate (VBR). In addition, realtime
applications (e.g. VoIP and IPTV) are delay-sensitive
while other non-real time services (e.g. FTP and Email) do not
need any guaranteed upper bound of the end-to-end delay.
Therefore, service providing for real-time variable bit rate (rtVBR)
sessions which require a bounded delay is more
challenging in comparison with other applications.
Different applications have different traffic natures which
imply different QoS requirements. QoS is usually quantified by
a single or combination of measurable parameters such as end-
978-1-4799-0801-1/12/$31.00 ©2012 IEEE
to-end delay, bandwidth, delay jitter (variation in delays) and
packet loss. To provide guaranteed QoS in a network, not only
the cooperation among different layers inside a network device
is required, but also these appliances should collaborate with
each other in the network in order to appropriately assign their
available resources to the demanding sessions . .In network
infrastructure, routers and data switches are the most important
devices in QoS provisioning. One of the components of a
switch/router that plays a significant role in providing QoS is
the scheduling module. A scheduling algorithm is a method
that chooses the most appropriate packet among competing
packets for transferring to a specified output port. A QoS
scheduling algorithm is the one that considers QoS parameters
in its decision making process. Thus, coincident with the
emergence of multimedia applications in recent years, many
researches have been conducted to achieve high performance
QoS scheduling algorithms on multimedia applications [1]-[6].
A well-known class of scheduling algorithms is Rate
Proportional Server (RPS) which isolates different sessions
from each other based on their service rates [7]. An ideal
scheduling algorithm in this class is Generalized Processor
Sharing (OPS) [8] which is used as a benchmark in evaluation
of practical algorithms. Some other types of packet-by-packet
scheduling disciplines such as SCFQ [9] and FFQ [10], try to
approximate OPS in real life.
When all the incoming traffic to the scheduler is constant
bit rate, an RPS server, say OPS, ideally share bandwidth
among them. But when an RPS server needs to schedule a
mixture of VBR and CBR flows, the provided QoS for rt-VBR
(bursty and delay-sensitive) sessions may be unacceptable. The
reason is that the packet delay in a VBR flow may be much
more than the packet delay in a CBR flow. We show this fact
by an example in section [I.