IOSR Journal of Computer Engineering (IOSRJCE) ISSN: 2278-0661, ISBN: 2278-8727 Volume 5, Issue 3 (Sep-Oct. 2012), PP 06-12 www.iosrjournals.

org

Delay Analysis and Optimality of Scheduling Policies For MultiHop Wireless Networks
1

A. Ashok kumar, 2K. Nageswara rao

Sathupally, MIST, Khammam.

Abstract: In this paper, we analyze the delay performance of a multi-hop wireless network in hich the routes
between source-destination pairs are fixed. We develop a new queue grouping technique to handle the complex correlations of the service process resulting from the multi-hop nature of the flows and their mutual sharing of the wireless medium. A general set based interference model is assumed that imposes constraints on links that can be served simultaneously at any given time. These interference constraints are used to obtain a fundamental lower bound on the delay performance of any scheduling policy for the system. We present a systematic methodology to derive such lower bounds. For a special wireless system, namely the clique, we design a policy that is sample path delay optimal. For the tandem queue network, where the delay optimal policy is known, the expected delay of the optimal policy numerically coincides with the lower bound. The lower bound analysis provides useful insights into the design and analysis of optimal or nearly optimal scheduling policies. We conduct extensive numerical studies to demonstrate that one can design policies whose average delay performance is close to the lower bound computed by the techniques presented in this paper.

I.

Introduction

A large number of studies on multi-hop wireless networks have been devoted to system stability while maximizing metrics like throughput or utility. These metrics measure the performance of a system over a long time-scale. For a large class of applications such as video or voice over IP, embedded network control and for system design; metrics like delay are of prime importance. The delay performance of wireless networks, however, has largely been an open problem. This problem is notoriously difficult even in the context of wireline networks, primarily because of the complex interactions in the network (e.g., superposition, routing, departure, etc.) that make its analysis amenable only in very special cases like the product form networks.bound and delay estimates for wireless networks with singlehop traffic were developed in [2]. However, the analysis is not directly applicable to multi-hop wireless network with multihop flows, due to the difficulty in characterizing the departure process at intermediate links. A typical multi-hop wireless network with multiple flows, each having exogenous arrivals at the source. Some of the important bottlenecks have been highlighted. source to its corresponding destination. The exogenous arrival processes AI (t), AII (t) and AIII (t) correspond to the number of packets injected in the system at time t. A packet is queued at each node in its path where it waits for an opportunity to be transmitted. Since the transmission medium is shared, concurrent transmissions can interfere with each others transmissions. The set of links that do not cause interference with each other can be scheduled simultaneously, and we call them activation vectors (matchings). We do not impose any a priori restriction on the set of allowed activation vectors, i.e., they can characterize any combinatorial interference model. For example, in a K-hop interference model, the links scheduled simultaneously are separated by at least K hops. each link has unit capacity; i.e., at most one packet can be transmitted in a slot. For the aboveexample, we assume a 1-hop interference model. The delay performance of any scheduling policy is primarily limited by the interference, which causes many bottlenecks to be formed in the network. we define a (K, X)-bottleneck to be a set of links X such that no more than K of them can simultaneously transmit. Figure 1 shows (1, X) bottlenecks for a network under the 1-hop interference model. In this paper, we develop new analytical techniques that focus on the queueing due to the (K, X)-bottlenecks. One of the techniques, which we call the reduction technique, simplifies the analysis of the queueing upstream of a (K, X)-bottleneck to the study of a single queue system with K servers as indicated in the figure. We now summarize our main contributions in this paper: Development of a new queue grouping technique to handle the complex correlations of the service process resulting from the multi-hop nature of the flows. We also introduce a novel concept of (K, X)-bottlenecks in the network. Development of a new technique to reduce the analysis of queueing upstream of a bottleneck to studying simple single queue systems. We derive sample path bounds on a group of queues upstream of a bottleneck. Derivation of a fundamental lower bound on the system wide average queuing delay of a packet in multihop wireless network, regardless of the scheduling policy used, by analyzing the single queue systems obtained above. www.iosrjournals.org 6|Page

Delay Analysis and Optimality of Scheduling Policies for Multi-Hop Wireless Networks
Extensive numerical studies and discussion on useful insights into the design of optimal or nearly optimal scheduling policies gained by the lower bound analysis. We begin with the description of the system model.

II. System Model We consider a wireless network G = (V,L), where V is the set of nodes and L is the set of links. Each link has unit capacity. There are N flows, each distinguished by its source destination pair (si, di). There is a fixed route (set of links) between the source si and corresponding destination di. Each route is a simple path. Each flow has its own exogenous arrival stream {Ai(t)}1 t=1. Each packet has a deterministic service time equal to one unit. The exogenous arrivals at each source are assumed to be independent. Let A(t) = (A1(t), . . . ,AN(t)) represent the vector of exogenous arrivals, where Ai(t) is the number of packets injected into the system by the source si during time slot t (for i 1, . . . ,N). Let _ = (_1, . . . , _N) represent the corresponding long-term average arrival rate vector. The path on which flow i is routed is specified as Pi := (v0 i , v1i , . . . , vj i , . . . , v|Pi| i), where vj i is a node at a j-hop distance from the source node si. The source node si is denoted by v0 i and the destination node di by v|Pi| i , where |Pi| is the path length. The packets arriving at each node are queued. i = 0. The queue length vector is denoted by Q(t) = (Qj i (t) : i {1, 2, . . . ,N} and j {1, 2, . . . , Pi}). Multiple flows can share a link e. i (t + 1) =_ Qj i (t) Ij i (t) + Ij1 i (t) if j > 0 Qj i (t) Ij i (t) + Ai(t) otherwise (II.2).

III.

Deriving Lower Bounds on Average Delay

In this section, we present our methodology to derive lower bounds on the system-wide average packet delay for a given. We use the indicator function 11{i2X} to indicate whether the flow i passes through the (K, X)-bottleneck, i.e., 11{i2X} = _1 if i X 0 otherwise. (III.3)The total flow rate _X crossing the bottleneck X is given by:_X = XN i=111{i2X}(_i). (III.4) Let the flow i enter the (K, X)-bottleneck at the node vki I and leave it at the node vli i . Hence, (liki) equals the number of links in the (K, X)-bottleneck that are used by flow i. We define _X and AX(t) as follows:_X = XN i=1 11{i2X}(li ki)(_i). (III.5) AX(t) = XN i=1 11{i2X}(li ki)(Ai(t)). (III.6) B. The reduction techniqueIn this section, we describe our methodology to derive lower bounds on the average size of the queues corresponding to the flows that pass through a (K, X)-bottleneck.By definition, the number of links/packets scheduled in the bottleneck, IX(t) is no more than K, i.e.,FX = N max i=1 11{i2X}(li ki). (III.8) Let Sk i (t) denote the sum of queue lengths of the first kqueues of flow i at time t, i.e., Sk i (t) = Xk j=0 Qj i (t) (III.9) SX(t + 1) = SX(t) IX(t) + AX(t). (III.12) Note: By summing the queues upstream of the bottleneck and defining SX(t), we are able to avoid correlation terms among the arrival and service processes in the queue evolution equation of the system (Eq.( III.12)). We obtain a lower bound on the value of SX(t) in Theorem 3.1 by studying a Reduced System: Consider a system with a single server and AX(t) as the input. The server serves at most K packets from the queue. Let QX(t) be the queue length of this system at time t. The queue evolution of the reduced system is givenby the following equation.QX(t + 1) = (QX(t) K)+ + AX(t) (III.10) where (x)+ = _ x if x > 0 0 otherwise The reduction procedure is illustrated in Fig. 2 where we have reduced one of the bottlenecks in the grid example. Proof: We prove the above theorem using the principle of mathematical induction. Base Case: The theorem holds true for T = 0, since the system is initially empty. Induction hypothesis: Assume that the theorem holds at a time T = t, i.e., QX(t) SX(t). Induction Step: The following two cases arise. Case 1: QX(t) K QX(t + 1) = QX(t) K + AX(t) SX(t) K + AX(t) SX(t) IX(t) + AX(t) = SX(t + 1). (III.14) Case 2: QX(t) < K. Using Eq. (III.11), we have the following, QX(t + 1) = AX(t) SX(t) IX(t) + AX(t) = SX(t + 1). (III.15) Hence, the theorem is holds for T = t + 1. Thus by the principle of mathematical induction, the theorem holds for all T. Remarks The above analysis captures the combinatorial interference constraints and reduces the bottleneck to a G/D/K system with appropriate inputs for the purpose of establishing lower bounds. Such a system can be analyzed for a large class of arrival traffic. Even when the arrival process is not amenable to analysis, the above reduction can be used to obtain a sample path lower bound via simulation. We emphasize that AX(t) can be computed from Eq. (III.6) and considers only the exogenous inputs to the system. Furthermore, the lower bound on the expected delay can be computed using only the statistics of the exogenous arrival process and not their sample paths.

www.iosrjournals.org

7|Page

Delay Analysis and Optimality of Scheduling Policies for Multi-Hop Wireless Networks
Bound on Expected Delay We now present a lower bound on the expected delay of the flows passing through the bottleneck as a simple function of the expected delay of the reduced system. In the analysis, we use Theorem 3.1 to bound the queueing upstream of the bottleneck and a simple bound on the queueing downstream of the bottleneck. Applying Littles law on the complete system, we derive a lower bound on the xpected delay of the flows passing through the bottleneck. Corollary 3.1: Let E[gDX] be the expected value of queuing delay for the G/D/1 system with input AX(t). Further let, E[DX] be the expected delay of the flows passing through X. Then E[DX] E[gDX] FX + XN i=1 11{i2X}_i(|Pi| li) _X Proof: Let Q(t) denote the queue length of the G/D/1 system at time t. Theorem 3.1 states that at all times, XN i=1 11{i2X} lXi1 j=ki Sj i (t) Q(t). Since for all j < li 1, Sj i (t) Sli1 i (t), thus XN i=1 11{i2X}(li ki)Sli1 i (t) Q(t). Using Eq. (III.8), it follows that, XN i=1 11{i2X}(FX)Sli1 i t) Q(t). (III.16) and hence, XN i=1 11{i2X} lXi1 j=0 Qj i Q(t) FX . (III.7) After crossing the bottleneck, a packet of flow i has to cross |Pi|li hops. Since the links are of unit capacity, the delay at each of these hops is at least one unit. Thus for all li j < |Pi|, E[Qj i ] _i. (III.18) Taking expectations on both sides of Eq. (III.17) and using Eq. (III.18), we obtain, XN i=1 11{i2X} X|Pi| j=0 E[Qj i ] E[Q(t)] FX + XN i=1 11{i2X}(_i(|Pi| li)). (III.19) Applying Littles law, E[DX] = XN i=1 11{i2X} lXi1 j=0 E[Qj i ] _X E[gDX] FX + XN i=1 11{i2X}_i(|Pi| li) _X. (III.20) Flow Partition Let Z be the set of flows in the system. Let _ be a partition on Z such that each element p _ is a set of flows passing through a common (Kp, Xp)-bottleneck. The expected delay of the flows in p can be lower bounded using Corollary 3.1. A system-wide lower bound on the expected delay of the packets, E[D], can then be obtained by a straight-forward application of the Littles law. E[D] X p2_ _XpE[gDXp ] FXp + XN i=1 11{i2X}_i(|Pi| li) PN i=1 _i (III.21) Our objective is to compute a partition _ such that the lower bound on E[D] can be maximized. Algorithm 1 Greedy Partitioning Algorithm 1: Z {1, 2 . . .N} 2: BOUND 0 3: repeat 4: Find the (K, X)-bottleneck which maximizes E[DX] 5: BOUND BOUND + _XE[DX] 6: Z Z \ i : i X 7: until Z = _ 8: return BOUND PN i=1 _i The (1, X)-bottlenecks correspond to cliques in the conflict graph [5]. constraints by grouping the arrivals into a single queue, and hence we underestimate the delay. We evaluate the impact of these relaxations on the accuracy of the lower bound using simulations. Despite these relaxations, we find that the lowerbound gives a useful estimate of the average delay in the system.

IV.

Design of Delay Efficient Policies

We now address the important question of designing adelay-efficient scheduler for general multi-hop wireless networks . Ensure high throughput: This is important because if the scheduling policy does not guarantee high throughput then the delay may become infinite under heavy loading. Allocate resources equitably: The network resources must be shared among the flows so as not to tarve some of the flows. C A D An example network (clique) with interference constraints such that only one pair of nodes can communicate at any given time. This is also known as sample path delay optimality. In particular, we will show that for the given network, scheduling the packet which is closest to its destination is optimal. Let h be the maximum number of hops, a flow takes in the clique network. h=N max i=1 |Pi| (IV.22)For the sake of simplicity, we define a h-dimensional vector q(t), which represents the state of the system: www.iosrjournals.org 8|Page

Delay Analysis and Optimality of Scheduling Policies for Multi-Hop Wireless Networks
q(t) =PN i=1 Q|Pi|1 i (t),PN i=1 Q|Pi|2 i (t), . . . , PN i=1 Q|Pi|h i (t)), where Qki = 0 k < 0. Note that we do not distinguish packets of different flows in this description of the state. Proof: The system is pre-emptive in that a different packet may be scheduled in the next slot. Lower Bound Since no more than one link can be scheduled in the clique network, it is a (1, X)-bottleneck. Let us define SX for the clique network (see Eq. III.11 of Section III-B). It can be shown that for the clique network, SX(t) =Xh k=1 kqk(t) (IV.25) We now show that any work-conserving policy (that schedules a non-empty queue) will achieve the lower bound on SX, i.e QX(t) at all times t. Suppose Ij is the activation vector scheduled by the policy. SX(t + 1) =Xj2 k=1 k(qk)(t) + (j 1)(qj1(t) + 1) + j(qj(t) 1) + AX(t)=Xh k=1 kqk(t) 1 + AX(t) = QX(t + 1) From the above we conclude that a policy which minimizes SX may not minimize the sum of queue lengths in the system at all times, neither it is guaranteed to be delay optimal. We give a formal description of the backpressure policy. Let e := (a, b) be a link of interest. Suppose that flow I passes through link e and that nodes a and b are at a distanceof j and j +1 hops, respectively, from the source node si. Inour notation, e := (vj i , vj+1 i ). Define the differential backlog Qi e of flow i passing through a link e := (vj i , vj+1 i ) as Qe i = (Qji i )_ (Qji+1 i )_, for some _ > 0 (IV.27) For each link e, the flow with the maximum differential backlog is chosen by the flow scheduling component (Eq. (IV.28) in Fig. 4). Flow Scheduling For each link e L, find the flow with the maximum differential backlog f_ e (t) = argmax i Qe i (IV.28) Assign weights to every link we = max( Qe f_ e , 0) (IV.29) Link Scheduling Schedule the maximum weighted matching I(t) = argmax www.iosrjournals.org 9|Page

Delay Analysis and Optimality of Scheduling Policies for Multi-Hop Wireless Networks
J2J hw, Ji (IV.3) where for two vectors x and y, hx, yi = P l xlyl denotes the inner product. Back-Pressure Policy With Fixed Routing Let kYk denote the Euclidean norm of vector Y. The system is considered to be stable if lim t!+1E[sup kQ(t)k] is bounded. ge =XN i=1 11{i2e}_i. (IV.31) Let g be the corresponding flow rate vector. It has been shown in [8] that if each arrival process is i.i.d. in time, and that the first two moments of all the arrival streams {Al(t)}1 t=1 are finite, then g C is a necessary condition for a stabilizing scheduling policy to exist. Arrival Processes: The arrival stream at each source is a series of active and idle periods. During the active periods, the source injects one packet into the queue in every time slot. The length of the active periods (denoted by random variable a)are distributed according the Zipf law with power exponent 1.25 and finite support [1,2,3,. . . ,100]. Truncated heavy-tailed distributions like Zipf, have been found to model the Internet traffic [7]. During the active period the source generates one packet every time-slot. The idle periods are geometrically distributed with mean p. (b) Dumbbell (c) Tree (d) Cycle III 100 80 50 40 35 30 18 1 A(t) sd Illustration of the Lower Bound analysis (bottlenecks used in the analysis have been highlighted). Simulation results for Tandem Queue Tandem Queue We consider a stream of packets flowing over the wireless links in tandem, as shown in Fig. 5(a) under the 1-hop interference model. For this system, any two links that are adjacent to mach other form an exclusive set. Choosing the first two links as the bottleneck maximizes the lower bound in Corollary 3.1 as it maximizes the value of (|Pi|li). Note that E[gDX] FX is the same for all exclusive sets in the system. The lower bound for the above arrival process is given by, C. Dumbbell topology Consider a dumbbell topology (Fig. 5(b)), with multiple flows passing through a single link with the load vector _(0.1, 0.2, 0.15, 0.05) packets/slot. The bottleneck exclusive set has been highlighted in Fig. 5(b). The performance of backpressure policy (_ = 0.1) is compared with that of the lower bound. System Utilization ( ) Expected Delay (slots/packet) 0 0.2 0.4 0.6 0.8 1 0 500 1000 1500 2000 FBFS BackPressure ( =1) BackPressure ( =0.5) BackPressure ( =0.1) Optimal Policy (LBFS) Lower Bound Simulation results for Clique System Load ( ) Expected Delay (slots/packet) 0 0.2 0.4 0.6 0.8 1 0 100 200 300 400 www.iosrjournals.org 10 | P a g e

Delay Analysis and Optimality of Scheduling Policies for Multi-Hop Wireless Networks
500 600 700 800 Maximal Scheduling BackPressure ( =0.1) The performance of backpressure policy (_ = 0.1) is significantly better than that of the maximal scheduling policy and is also close to the lower bound. E. Cycle Topology We now illustrate the application of the lower bound analysis to a (2, X)-bottleneck. For the given cycle network in Fig. 5(d), no more than two links can be scheduled at any time under 2-hop interference constraints, i.e. for X = {1, 2, 3 . . . , 8}, IX(t) 2. It can be easily verified that the analysis presented I n Section III-B can be used to reduce the A grid network with multiple flows. Some of the important bottlenecks have been highlighted. System to a G/D/2 queue having arrivals 4AI (t) and 4AII (t) respectively. We studied the system for several different input load vectors _. We find that depending on theinput load vector, Algorithm 1 computes different partitions for the flows in the system. We discuss two representativeload vectors to evaluate the impact of the relaxations made in the analysis. Case 1: (0.12, 0.15, 0.12, 0.06, 0.12, 0.15, 0.12) packets/slot. For the given load vector, Algorithm 1 computes the partition{{II, VI};{I, V};{III};{IV};{VII}}. Note that flow IV interferes with flows II and VI significantly, but this effect is not captured by the lower bound analysis. Also note that flow I interferes with flows IV and VI. The lower bound computedby Algorithm 1 is 185.9 slots/packet. System Load ( ) Expected Delay (slots/packet) 0 0.2 0.4 0.6 0.8 1 0 100 200 300 400 500 Maximal Scheduling BackPressure ( =0.1) Lower Bound Fig. 10. Simulation results for Tree Topology System Load Expected Delay (slots/packet) 0.2 0.4 0.6 0.8 1 50 100 150 200 250 300 350 400 450 BackPressure ( =0.25) BackPressure ( =0.5) Lower Bound (K=2) Lower Bound (K=1) Fig. 11. Simulation results for Cycle Topology values of _. The average delay was found to be 308.0 slots/packet for _ = 0.4. For smaller values of _, flow VI was starved for resources resulting in larger delays. The average delay for _ = 0.25 and _ = 0.1 was 323.3 slots/packet and 476.2 slots/packet respectively. Case 2: _ = (0.12, 0.15, 0.12, 0.0, 0.12, 0.15, 0.12) packets/slot. In this case, we remove flow IV from the system and keepall other arrivals rate the same. The lower bound computed by Algorithm 1 is 196.0 slots/packet.

www.iosrjournals.org

11 | P a g e

Delay Analysis and Optimality of Scheduling Policies for Multi-Hop Wireless Networks
G. Linear Network Finally, we discuss an example of a line network with several short flows and a single long flow as shown in Fig. 12. The packet arrival rate (_) is the same for all the flows I II III IV V VI VII VIII IX X XI 1 2 3 4 5 6 7 8 9 10 11Fig. 12. A linear network with multiple short flows and a single long flow. Arrival rate ( )
Expected Delay (slots/packet) 0.1 0.15 0.2 0.25 0 20 40 60 80 100 120 140 160 BP ( =1) BP ( =0.1) Shadow ( =1) Shadow ( =0.1) New Policy

Lower Bound

V.

Conclusion

`The delay analysis of wireless networks is largely an open problem. In fact, even in the wireline setting, obtaining analytical results on the delay beyond the product form types of networks has posed great challenges. These are further exacerbated in the wireless setting due to complexity of scheduling needed to mitigate interference. Thus, new approaches are required to address the delay problem in multi-hop wireless systems. To this end, we develop a new approach to reduce the bottlenecks in a multi-hop wireless to single queue systems to carry out lower bound analysis. For a special class of wireless systems (cliques), we are able to apply known techniques to obtain a sample path delay-optimal scheduling policy. We also obtain policies that minimize a function of queue lengths at all times on a sample path basis. Further, for a tandem queueing system, we show numerically that the expected delay of a previously known delay-optimal policy coincides with the lower bound. The analysis is very general and admits a large class of arrival processes. Also, the analysis can be readily extended to handle channel variations. The main difficulty, however is in identifying the bottlenecks in the system. The lower bound not only helps us identify near-optimal policies, but may also help in the design of a delay-efficient policy as indicated by the numerical studies.

References:
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] J. G. Dai and W. Lin. Asymptotic optimality of maximum pressure policies in stochastic processing networks. Preprint, October 2007. H. Dupuis and B. Hajek. A simple formula for mean multiplexing delayfor indep endent regenerative sources. Queueing Systems Theory and Applications, 16:195239, 1994. A. Feldmann, N. Kammenhuber, O. Maennel, B. Maggs, R. D. Prisco, and R. sundaram. A methodology for estimating interdomain web traffic demand. In IMC, 2004. L. Georgiadis, M. J. Neely, and L. Tassiulas. Resource Allocation andCross-Layer Control in Wireless networks, Foundations and Trends in Networking, volume 1. Now Publishers, 2006. G. R. Gupta. Delay Efficient Control Policies for Wireless Networks.Ph.D. Dissertation, Purdue University, 2009. G. R. Gupta, S. Sanghavi, and N. B. Shroff. Node weighted scheduling.SIGMETRICS-Performance09, June 2009. G. R. Gupta, S. Sanghavi, and N. B. Shroff. orkload optimality in switches without arrivals. MAthematical performance Modeling and Analysis Workshop, june 2009. G. R. Gupta and N. B. Shroff. Delay analysis for wireless networks with single hop traffic and general interference constraints. IEEE Transactions on Networking, 18:393 405, April 2010. I. Ilog. Solver cplex, 2007. http://www.ilog.com/products/cplex/. S. Jagabathula and D. Shah. Optimal delay scheduling in networks with arbitrary constraints. In ACM SGMETRIC/Performance, June 2008. K. Jain, J. Padhye, V. Padmanabhan, and L. Qiu. Impact of interference on multi-hop wireless network performance.InMOBICOM, 2003

A.Ashok Kumar pursuing M.Tech in the department of Computer Science & Engineering. His Interested areas are Computer networks, data mining and data warehousing. K.Nageswara Rao working as a associate professor in department of Computer Science & Engineering. His Interested areas Computer networks, data mining and data warehousing.

www.iosrjournals.org

12 | P a g e