Sie sind auf Seite 1von 6

International Journal of Wireless Communications and Networking 3(1), 2011, pp.

39-44

DELAY ANALYSIS OF WIRELESS NETWORK USING MWM


SCHEDULING POLICY
Kiruba, J. and V.Thiruppathy Kesavan
Dept. of CSE, Kalasalingam University, Krishnankoil, India, E-mail: jkiruba12@gmail.com

Abstract: The paper proposes the solution to the wireless networks where the delay analysis of wireless
networks with general interference constraints on the set of links that can be served simultaneously at any
given time systems has been limited to asymptotic behavior in the heavy traffic regime and order results.
The single-hop traffic allows for simultaneous transmissions as long as they satisfy the underlying interference
constraint which is proved by finding a lower bound on the delay performance of any scheduling scheme
for this system. The paper aims at developing an estimate for the expected delay of wireless networks with
mutually independent arrival streams operating under the well-known maximum weighted matching
(MWM) scheduling policy. The simulation results shows that the delay performance of the MWM policy is
often close to the lower bound, which means that it is not only throughput optimal, but also provides excellent
delay performance.
Keywords: Maximum Weighted Scheduling, lower bound, upper bound.

I. INTRODUCTION including datagrams and virtual circuits.


The important aspect of scheduling wireless data Furthermore the time varying nature of a wireless
in systems such as 802.16 (WIMAX) problems is that network, due either to fading channels or to
a scheduler knows the channel rates across all users changing connectivity due to mobility, is adequately
and all carriers whenever a scheduling decision is captured in the model to allow for state dependent
made. Hence there is no need to treat each carrier network control policies [3].
in complete isolation. This gives a potential for The scheduling algorithms that do not consider
enhancing performance by allocating multiple queue backlog must incur average delay that grows
carriers simultaneously. In a situation where finite at least linearly with N. A dynamic queue-length
queues are fed by a data arrival process, the aware algorithm maximizes throughput and
MaxWeight algorithm can be used for the single- achieves an average delay that is independent of
carrier setting to accommodate a number of natural N [4]. This interference between links significantly
optimization problems in the multi-carrier setting[1]. complicates the scheduling component in wireless
When arrivals are modulated by independent finite systems. This is especially problematic in multi-hop
state Markov chains, the maximal scheduling wireless networks, where the number of nodes can
algorithm achieves average delay that grows at be quite large and one cannot assume the presence
most logarithmically in the largest number of of a central authority. The extent of the interference
interferers at any link. When each Markov process varies across networks as it is physical layer
has at most two states (such as bursty ON/OFF dependent, and with it varies the complexity of
sources), average delay is independent of the scheduling [5].
number of nodes and links in the network, and A class of approximation algorithms to MWM
hence is order-optimal [2]. called 1-APRX which always obtain a schedule
A model that captures the cross-layer whose weight W differs from the weight of MWM
interaction from the physical to transport layer in schedule W! by at most f(W!), where f(.) is a sub-
wireless network architectures including cellular, linear function is presented. This difference in
ad-hoc and sensor networks as well as hybrid weight is referred to as “approximation distance”
wireless-wireline allows for arbitrary network of algorithm from MWM. Any 1-APRX algorithm
topologies as well as traffic forwarding modes, is stable, that is, it delivers upto 100% of
40 International Journal of Wireless Communications and Networking

throughput under any admissible Bernoulli i.i.d. transport layer and the link layer due to the fact
input traffic [6]. that the attainable throughput on each link in the
The total network throughput of the queue- network is ‘elastic’ and is typically a non-convex
length-based policy is no less than that of the greedy and non-separable function of the transmission
policy for all N and is strictly larger than the attempt rates. Two cross-layer algorithms, a dual
throughput of the greedy policy for large N. Further, based algorithm and a primal based algorithm, are
given an upper bound on the delay violation proposed to solve the rate control problem in a
probability, the throughput of the queue-length- multi-hop random access network. Both algorithms
based policy is an increasing function of N while the can be implemented in a distributed manner, and
throughput of the greedy policy eventually decreases work at the link layer to adjust link attempt
with increasing N and goes to zero. Given an upper probabilities and at the transport layer to adjust
bound on the queue overflow probability, the session rates [14]. The performance of cross-layer
throughput of the queue-length-based policy is a congestion control will be impacted if the network
strictly increasing function of N while the throughput can only use an imperfect and potentially
of the greedy policy eventually goes to a constant [7]. distributed scheduling component that is easier to
implement [15].
Using ideas from the Lyapunov function
approach in control theory, this technique maps the Internet routers frequently use a crossbar switch
complex multi-dimensional calculus of variations to interconnect line cards. The crossbar switch is
problem to a one-dimensional calculus of variations scheduled using an algorithm that picks a new
problem, and the latter is often much easier to solve. crossbar configuration every cycle. Several
This technique can potentially be used to study the scheduling algorithms have been shown to
delay-performance of a large class of wireless guarantee 100% throughput under a variety of
scheduling algorithms [8]. When the overflow metric traffic patterns. The first such algorithm was the
is appropriately modified, the minimum-cost-to- maximum weight matching (MWM) algorithm in
overflow under the α-algorithm can be achieved by which the weight is the sum of the occupancies of
a simple linear path, and it can be written as the the queues. Alternative weight functions, such as
solution of a vector-optimization problem [9]. using the sum of the square of the occupancies can
be used which leads to stronger or weaker stability.
The capacity (service rate) of the channel varies
randomly with time and asynchronously for
II. EXISTING SYSTEM
different users [10]. Packets randomly enter the
system at each node and wait in output queues to In a wireless system, many scheduling policies have
be transmitted through the network to their been studied at the MAC layer with the objective
destinations. A joint routing and power allocation of maximizing throughput. These schemes are often
policy is developed which stabilizes the system and called throughput-optimal scheduling schemes.
provides bounded average delay guarantees This paper derives the upper and lower bounds on
whenever the input rates are within this capacity the expected delay, and also provides an accurate
region. Such performance holds for general arrival estimate of the expected delay for a well-known and
and channel state processes, even if these processes extensively studied throughput-optimal scheme
are unknown to the network controller [11]. An called the maximum weighted matching (MWM).
optimization approach to flow control is to To simplify the analysis, the traffic model is
maximize the aggregate source utility over their restricted to single-hop traffic. Under the single-hop
transmission rates [12]. traffic model, all packets transmitted on a link are
In Quasi Unit Disk Graphs, flooding is an generated by an exogenous arrival process at the
asymptotically message-optimal routing technique source node. The design of scheduling policies
which provide a geometric routing algorithm being which stabilize the system even under single-hop
more efficient above all in dense networks, and traffic is a challenging task.
classic geometric routing is possible with the same Intuitively, the scheduler must schedule as
performance guarantees as for Unit Disk Graphs if many links as possible in every time slot, such
d ≥1/√2 [13]. schedulers are called maximal schedulers.
The rate control in a multi-hop random access However, even with maximal scheduling, some of
network requires joint optimization at both the the queue lengths may become unbounded. The
Delay Analysis of Wireless Network using MWM Scheduling Policy 41

reason is that if the scheduler does not use the queue noted that the MWM policy achieves load balancing
length information, some of the queues may grow without explicit knowledge of the arrival statistics,
large, while others remain very small or become simply by using the information of the backlogs and
empty. This, in turn, does not allow the scheduler thus achieves a delay performance comparable to
to schedule a large number of queues and leads to that of the GMWMopt.
instability. Thus, a throughput optimal policy like
MWM carefully uses the information of the queue A. Computation of Lower Bound
lengths while scheduling the links.
This paper proposes an algorithm to calculate a
lower bound on the delay of the system,
III. PROPOSED SYSTEM independent of the scheduling policy used. The
The paper proposes that the behavior caused by definition of the exclusive sets, of link in the system
throughput-efficient schedulers significantly is that only one of the queues in can be scheduled
complicates the delay analysis of these systems, at any given time slot. The notion of exclusive sets
because the service process of each link is governed is helpful for deriving fundamental lower bounds
not only by the interference constraints, but also by on the expected delay of the system. The value of
its queue length. For example, in a wireless network the lower bound is incremented and the links in the
operating under a throughput optimal policy, such chosen exclusive set are removed from further
as the MWM policy, the expected delay at a link consideration. This process is repeated until every
may be large even if the arrival rate is small. This is link in the system has been used. Since each link
because these policies try to schedule the longer appears in exactly one exclusive set, the system-
queues in the system or in other words, they prevent wide lower bound on the expected queue length can
the queues from becoming very large. This can be be obtained as the sum of the contribution of each
thought of as a mechanism to balance the queue link toward the lower bound. Consider X be the set
lengths in the system. of N number of Systems connected in the Network.
The main contribution of this paper is two- fold. Step 1: Initially set a bound value to 0. This
First, a fundamental lower bound is developed bound value is used for storing the Lower bound
based on the expected queuing delay of a wireless value in the extreme subset X1 from the set X.
network regardless of the scheduling policy used. Step 2: Take exclusive subset X1 from the set X,
Then, an upper bound based on the expected delay which maximize the lower bound value.
of a throughput for the proposed optimal
scheduling policy called GMWM which is a Step 3: Calculate the bound value by adding the
generalization of MWM is developed under a bound and lower bound value of the exclusive
single-hop traffic model. Secondly, an estimate for subset X1.
the expected delay in a wireless network with Step 4: Divide the entire number of system
mutually independent arrival streams, under a connected in the network by the subset of the system.
MWM type policy, given the load and the
interference constraints is developed. Further, the Step 5: Repeat the steps 3, 4 and 5, until the X is
estimate is shown to lie between the upper and equal to null.
lower bounds. Through simulations it is shown that Step 6: Return the Bound value.
for single-hop traffic and any given load within the
capacity region, the estimate is accurate. B. Minimization of Upper Bound
This paper shows that the MWM type of policies This paper analyses a class of generalized maximum
which were designed primarily for achieving weighted matching policies, parameterized by
maximum throughput, indeed also have good delay Weights. The MWM policy is a special case, where
performance. This can be attributed to two reasons. all the weights are unity. We prove that GMWM
First, MWM schedules a maximal set of links in the achieves 100% throughput using the Foster-
system. Second, it performs load balancing in the Lyapunov drift criteria for Countable Markov
system. chains. The Algorithm to minimize the upper bound
In this paper the impact of GMWM type of by computing the network utility maximization
scheduling policies on the expected queue lengths using convex optimization techniques is stated
and expected delay in the system is analyzed. It is below. Let a be the set of prices and ç be a variable.
42 International Journal of Wireless Communications and Networking

Step 1: First initialize a to the power of η.


Step 2: Optimal value µ is calculated by using
the queue length value λ and the price of the set a.
Step 3: The sum of optimal value for each subset
from the entire set is calculated and the each subset
upper bound value is minimized.
Step 4: Increment the price value for calculating
the upper bound value for the entire network.
Step 5: Repeat the steps 3, 4 and 5, till the
optimal value.
Step 6: Return the boundary value.

C. Estimation of the Delay


This paper analyses the delay of the wireless
network when operated with a stationary Figure 1: Packet Delay under Uniform Traffic with Packet
randomized scheduler. The service process can be Size as 1 Cell
analyzed as follows. The scheduler is unaware of
the backlog and chooses to schedule link Under uniform traffic condition three cases are
independent of whether the queue is empty or not. considered.
In every slot, if the link is scheduled, exactly one Pattern 1: packet size is 1 cell.
packet is served, otherwise the packets in the queue Pattern 2: packet size is 10 cells.
wait for the next available slot. The following
Pattern 3: packet size is varied and the average
parameters must be defined for the system.
is 10 cells (Internet packet size distribution).
• Length of the queue at the beginning of time The average packet delay is considered for
slot. various schemes for different patterns in figure 1-3.
• Number of arrivals at link during the time Under non-uniform traffic condition two cases
slot. are considered.
• Random variable that is 1 if link is Pattern 1: Diagonal traffic, packet size is 1 cell.
scheduled and is 0 otherwise.
Pattern 2: Hotspot traffic, packet size is 1 cell.
• Average delay in the system.
The lower bound may not be achieved by any
policy because it may not be possible to schedule a
link in every exclusive set due to the interference
constraints. Therefore, we attempt to develop an
accurate estimate for the delay performance in this
section. The lower bound analysis suggests that
those exclusive sets that have a large must have
longer queues lengths. However, since a scheduling
policy like MWM also balances the queue lengths
in the system, the effect of congestion in a particular
exclusive set is distributed over the whole system.
Hence, instead of estimating the queue length at
each link, we estimate the contribution of each link
toward the aggregate expected queue length.

IV. SIMULATION RESULTS


The Simulation results are made by considering Figure 2: Packet Delay under Uniform Traffic with Packet
both the uniform and non-uniform traffic patterns. Size as 10 Cells
Delay Analysis of Wireless Network using MWM Scheduling Policy 43

Figure 5: Reassembly Delay under Uniform Traffic with


Packet Length Larger than 1 Cell
Figure 3: Packet Delay under Uniform Traffic with Variable
Packet Length and the Average Length being10 Table I
Cells Summary of Performance of Various Algorithms
Schemes Complexity Stable Packet Delay
Performance

HE-iSLIP O(logN) Yes Lowest when


packet size is
larger than 1 cell.
iSLIP O(logN) No Always higher
than HE-iSLIP.
DERAND O(logN) Yes Highest for all
traffic patterns.
SERENA O(N) Yes Lower than
HE-iSLIP only
under nonuniform
diagonal traffic.
MWM O(N3) Yes Lowest when
packet size is 1 cell.

V. CONCLUSION
Figure 4: Cell Delay under Uniform Traffic with Packet
Length Larger than 1 Cell The lower bound may not be achieved by any policy
because it may not be possible to schedule a link in
every exclusive set due to the interference
constraints. Therefore, we attempt to develop an
When Packet Length is Larger than 1 Cell, HE- accurate estimate for the delay performance in this
iSLIP have a lower packet delay than MWM. For section. The lower bound analysis suggests that
example, when packet length is 10 cells, the Cell those exclusive sets that have a large must have
delay obtained is shown in Figure 4 and the longer queues lengths. However, since a scheduling
Reassembly delay is shown in Figure 5. It is policy like MWM also balances the queue lengths
concluded that low cell delay and low reassembly in the system, the effect of congestion in a particular
delay are needed for low packet delay. exclusive set is distributed over the whole system.
Hence, instead of estimating the queue length at
The performance of various schemes in terms each link, the contribution of each link toward the
of Complexity, stability and packet delay are aggregate expected queue length is estimated in this
compared and shown in table 1. paper.
44 International Journal of Wireless Communications and Networking

ACKNOWLEDGMENT Allerton Conf. on Communication, Control, and


Computing, 2006.
The authors wish to thank all the reviewers whose
comments have helped to bring this research paper. [9] V. J. Venkataramanan and X. Lin, “Structural
properties of ldp for queue-length based wireless
REFERENCES scheduling algorithms,” 44th Annual Allerton Conf.
on Communication, Control, and Computing, Sep.
[1] M. Andrews and L. Zhang, “Scheduling algorithms 2006.
for multi-carrier wireless data systems,” in Proc. [10] S. Shakkottai, R. Srikant, and A. Stolyar, “Pathwise
MobiCom, 2007, pp. 3–14. optimality of theexponential scheduling rule for
[2] M. J. Neely, “Delay analysis for maximal scheduling wireless channels,” Advanced Applications of
in wireless networks with bursty traffic,” in Proc. of Probability, Dec. 2004.
IEEE INFOCOM, 2008, pp. 6–10. [11] M. J. Neely, E. Modiano, and C. E. Rohrs, “Dynamic
[3] L. Georgiadis, M. J. Neely, and L. Tassiulas, power allocation and routing for time varying
“Resource Allocation and Cross-Layer Control in wireless networks,” IEEE Journal on Selected. Areas
Wireless Networks, Foundations and Trends in Communication, Vol. 23, No. 1, pp. 89–103, 2005.
inNetworking” New York: Nov, 2006, Vol. 1. [12] S. H. Low and D. E. Lapsley, “Optimization flow
[4] M. J. Neely, “Order optimal delay for opportunistic control: Basic algorithm and convergence,” IEEE/
scheduling in multi-user wireless uplinks and ACM Trans. on Networking, Vol. 7, No. 6, pp. 861–
downlinks,” 44 th Annual Allerton Conf. on 874, 1999.
Communication, Control, and Computing, Sep. [13] F. Kuhn, R. Wattenhofer, and A. Zollinger, “Ad-hoc
2006. networks beyond unit disk graphs,” in Proc. 1st
ACM Joint Workshop on Foundation on. Mobile
[5] G. Sharma, C. Joo, and N. B. Shroff, “Distributed
Computing (DIALM-POMC), San Diego, CA, Sep.
scheduling schemes for throughput guarantees in
2003, pp. 69–78.
wireless networks,” 44th Annual Allerton Conf. on
Communication, Control, and Computing, Sep. [14] X. Wang and K. Kar, “Cross-layer rate control for
2006. end-to-end proportional fairness in wireless
networks with random access,” in Proc. MobiHoc,
[6] D. Shah and M. Kopikare, “Delay bounds for the 2005, pp. 157–168.
approximate maximum weight matching algorithm
[15] X. Lin and N. B. Shroff, “The impact of imperfect
for input queued switches,” in Proc. of IEEE
scheduling on cross-layer congestion control in
INFOCOM, 2002, Vol. 2, pp. 1024–1031.
wireless networks,” IEEE/ACM Trans. on
[7] L. Ying, R. Srikant, A. Eryilmaz, and G. E. Dullerud, Networking, Vol. 14, No. 2, pp. 302–315, 2006.
“A large deviations analysis of scheduling in [16] Keslassy and N. McKeown, “Analysis of scheduling
wireless networks,” IEEE Transactions on Information algorithms that provide 100% throughput in input-
Theory, Vol. 52, No. 11, pp. 5088–5098, 2006. queued switches,” presented at the 39th Annual
[8] X. Lin, “On characterizing the delay performance Allerton Conf. on Communication, Control, and
of wireless scheduling algorithms,” 44th Annual Computing, 2001.

Das könnte Ihnen auch gefallen