Docstoc

paper-8

Document Sample
paper-8 Powered By Docstoc
					                                                              International Journal of Computer Information Systems,
                                                                                                   Vol. 5, No.2, 2012

             An Integrated Approach To Optimizing
             Performance In Transaction Network
                     Serving Mixed Flows
                       1                                                              2
                           A.RARAJU                                                       K.SRINIVAS
    1                                                                             2
     M.Tech (CSE) Kottam College of Engineering,                                 Professor and HOD,
            Kurnool. raraju22@gmail.com                                    Kottam College of Engineering,
                                                                            Kurnool, kipgs2008@gmail.com

    Abstract—In this work, we study the control of                  structure of the solution, and forces us to develop
    communication networks in the presence of both                  a joint congestion control and load balancing
    inelastic and elastic traffic flows. The                        mechanism that is fully distributed, and achieves
    characteristics of these two types of traffic differ            high throughput and good delay characteristics.
    significantly. Hence, earlier approaches that focus
    on homogeneous scenarios with a single traffic type
                                                                    Our goal is to balance the load of the inelastic
    are not directly applicable. We formulate a new                 traffic in the network such that the elastic traffic
    network optimization problem that incorporates                  intelligently exploits the time varying residual
    the performance requirements of inelastic and                   capacity (the link capacity minus the capacity
    elastic traffic flows. The solution of this problem             needed to serve the inelastic flows) at each link in
    provides us with a new queueing architecture, and               the network. To see the potential gains of such an
    distributed load balancing and congestion control               interaction, consider the network shown in Figure
    algorithm with provably optimal performance. In                 1, which serves one inelastic and one elastic flow
    particular, we show that our algorithm achieves the             over links of capacity 20. Assume that the
    dual goal of maximizing the aggregate utility
    gained by the elastic flows while satisfying the
                                                                    inelastic flow has a fixed rate of 20 and has two
    demands of inelastic flows. Our base optimal                    routes to divide its traffic over as shown in the
    algorithm is extended to provide better delay                   figure. It can be seen that the rate distribution
    performance for both types of traffic with minimal              decision of the inelastic flow will significantly
    degradation in throughput. It is also extended to               affect the elastic flow performance. If it divides
    the practically relevant case of dynamic arrivals               its traffic equally amongst the two routes as in
    and departures. Our solution allows for a                       Figure 1, the elastic flow cannot achieve a rate
    controlled interaction between the performance of               more than 10. However, if the inelastic flow can
    inelastic and elastic traffic flows. This performance           steer more of its traffic over the less congested
    can be tuned to achieve the appropriate design
    tradeoff. The network performance is studied both
                                                                    route, more resources become available to the
    theoretically and through extensive simulations.                elastic traffic and it can achieve rates close to 20
                                                                    as shown in Figure 2. With this intuition, we want
    I. INTRODUCTION                                                 to design a dynamic algorithm that automatically
    Over the last several years, we have witnessed the              adapts the operation of inelastic and elastic flows
    development of increasingly sophisticated                       to get the optimal performance. This requires a
    optimization and control techniques to address a                solution that seamlessly and distributively
    variety of resource allocation problems for                     balances the load of the inelastic traffic across the
    communication networks (e.g. [2], [11], [19], [1],              network as well as injects enough elastic traffic
    [10], [15], [23], [6], [21], [26], [13], [7], see [17],         into the network so that no capacity is wasted
    [8] for an overview). Much of this investigation                while preventing network overloading.
    has focused primarily on controllable or elastic                Fig.
    traffic. Thus, it is imperative that one develop
    efficient resource allocation strategies to jointly
    manage both inelastic and elastic traffic.
    Integration of elastic and inelastic flows in single-
    hop wireless systems has been studied [24], [3],
    [22], and in [9] it has been extended to a
    multiple-hop network, however with the                          Fig:1.Fixed inelastic Traffic
    restriction of every flow having a single route.                Fig:2.Controllable inelastic Traffic
    The availability of multiple routes, which is
    studied in this paper, significantly changes the



August Issue                                            Page 48 of 68                                      ISSN 2229 5208
                                                            International Journal of Computer Information Systems,
                                                                                                 Vol. 5, No.2, 2012

    We begin first by providing the system model                  [t])rЄRi ,we note that Ai[t] denotes the number of
    and general assumptions. We then formulate a                  packets generated by flow fi while (xi(r)[t])rЄRi
    problem that attempts to maximize the utility of              describes the number of packets injected into the
    the elastic flows in the network subject to the               network to traverse each of the available routes of
    constraint that the data requirements of the                  flow fi. Thus, Ai[t] is an uncontrollable stochastic
    inelastic flows are met. We solve this problem via            process describing exogenous arrivals, whereas
    a two-step approach. First, we solve a simple                 (xi(r)[t])rЄRi is controllable by the network
    version of the problem when the inelastic flow                algorithm. For notational convenience, we define
    rates are deterministic. We then use the insights
    gained from that framework and extend the
    solution to the more general stochastic case. We
    then extend the work in two practically important
    directions. The first is to develop a virtual queue           to denote the total number of inelastic packets on
    based solution that allows us to achieve low                  link l for a given xi[t]. Elastic Flow: We let fe
    delays with a nominal and controllable sacrifice              denote an elastic flow in the network with source
    in the throughput of the elastic flows. The second            se and destination de. We assume that each
    is to extend the solution in the presence of flow             elastic flow fe is associated with a single route
    arrivals and departures, where certain elastic                Re; and we let xe[t] be the the number of injected
    flows may be very short and may leave the                     packets of flow fe in slot t: Similar to the inelastic
    system before the algorithm has the opportunity               case, we also define xe[t]:=(xe[t])feЄFe to be the
    to converge. We also present extensive                        vector of elastic flow rates in slot t; and
    simulations to demonstrate the interaction
    between the two types of flows under our
    proposed algorithms. In particular, we show that
    due to the dynamic nature of the load balancing               to denote the total number of elastic packets on
    mechanism implemented by inelastic sources, the               link l. Associated with each elastic flow fe there
    elastic flows are able to push inelastic traffic onto         exists a utility function Ue(.).that measures the
    less loaded routes and achieve higher rates. We               “satisfaction” of that flow as a function of its
    show that this interaction maximizes the sum of               mean                                     injection
    the utilities of the elastic flows while satisfying           rate
    the demands of the inelastic flows. We also                   In the text, we use x[t] := (xi[t],xe[t]) to denote
    compare the delay performance of our algorithm                the vector of inelastic and elastic packets injected
    with and without the virtual queue                            into the network in slot t: Next, we provide a set
    implementation and illustrate that the virtual                of assumptions to be used later in the analysis:
    queue scheme can reduce the end-to-end delays                 Assumption 1:The elastic routing matrix
    significantly.                                                [Re]feЄFe has full row rank, which guarantees
    II OBJECTIVES: Inelastic Flow: We let fi denote               that given q; there exists a unique p such that q =(
    an inelastic flow in the network with source si               [Re]feЄFe )Tp.
    and destination di. Each inelastic flow fi is                 Assumption 2:The inelastic arrival process
    associated with a fixed set of routes Ri. The rth             {Ai[t]}fi Є Fi is such that there exists a vector xi
    route of this set is described by a vector Ri(r)
    such that Ri(r) [l] = 1.if link lЄL is on that route,
    and zero otherwise. Let xi(r)[t] be the number of
    injected packets on the rth route of flow fi at time          satisfying                                        and
    slot t, and let xi[t] := (x i (r)[t]R(r) ЄRi fiЄFi be
    the vector of inelastic flow packets injected on
    each route in slot t: Note that we slightly abuse             This condition implies that the inelastic flows are
    our notation by using xi to denote rate vector of             supportable by the network, i.e., there exists a
    all inelastic flows, while i(r)stands for the rate of         rate division of the inelastic flow rates over their
    flow fiЄFi over route rЄRi. We assume that the                available routes which can support the arriving
    packet arrivals of the inelastic flow fi follow a             traffic.
    stochastic process Ai[t] that is identically and              Assumption 3: The utility functions {Ue(xe)}fe
    independently distributed (i.i.d.) over time with a           are strictly concave, twice differentiable, and
    fixed mean rate, denoted by ai := E(Ai[t]), and a             increasing functions. Such an assumption is
    finite second moment, i.e. E(A2i [t]) < ∞.To                  commonly used to capture the diminishing
    clarify the difference between Ai[t] and (x i (r)




August Issue                                          Page 49 of 68                                       ISSN 2229 5208
                                                            International Journal of Computer Information Systems,
                                                                                                 Vol. 5, No.2, 2012

    returns to the elastic flows of an increase in the            evolution in (1) which possess a more tractable
    service rate.                                                 and cleaner form.
    Assumption 4: For each elastic flow feЄFe, its                Definition 1 (Stability): We say that a queue qR
    utility function Ue(x) satisfies: for each m>0 and            is stable if
    MЄ[m, ∞],there exists                   and with
                                                                                                                (2)
    ,                           ,satisfying                       where B is some finite positive value. We say
                                                                  that the network is stable if all aggregate queues
    for all xЄ[m,M].We note that Assumption 1 is                  {qR} for both inelastic and elastic flows are
    not critical in the proof of stability but will               stable.
    simplify our proof. Also note that Assumptions 3              III. JOINT CONGESTION CONTROL AND
    and 4 on the utility functions are not restrictive            LOAD BALANCING Stochastic Network
    and hold for the following class of utility                   Optimization (SNO) Problem:
    functions U(x)=wx(1-a) /(1-a),for a>0,which is
    known to characterize a large class of fairness
    concepts such as max-min fairness and weighted                                                      (3)
    proportional fairness ([25] and the references                s.t. The Queue Evolution as in(1),Network
    therein).In subsequent discussions, when the                  Stability           as               in(2),
    distinction between real and non-real-time routes
    is unnecessary, we will simply refer to a route as
    R without any subscripts. Furthermore, for
    simplicity, we will use zl[t] for zl(xi[t]) and yl[t]         A. Heuristic Fluid Model
    for     yl(xe[t]).Queueing     Architecture      and          In the fluid model scenario, all the dynamics and
    Evolution: In our system, for each link (I,j) Є L; a          randomness are ignored, and the stochastic
    single priority queue is maintained at the                    constraints are replaced with static constraints. In
    transmitting node i, which holds all the packets              particular, the inelastic flow fi is assumed to have
    whose routes traverse (I,j): Since the inelastic              a fixed arrival rate ai, and the network stability
    flows are expected to have more stringent delay               condition is replaced by a condition on total link
    constraints, their packets are always stored ahead            rate being no more than capacity. Then, the SNO
    of those of the elastic flows. We let pl[t] denote            problem reduces to the following problem in this
    the queue length of the buffer associated with link           scenario.
    l at the beginning of slot t, and define                      Fluid Network Optimization (FNO) Problem:


    to be the total queue length on route R. Notice
    that pl[t] and qR[t] counts both the inelastic and
    elastic flows’ packets. During each time slot, the
    queue            pl          evolves            as


    where x+ = max(0, x). This evolution is based on              In our discussion, we will abbreviate the
                                                                  aggregate elastic and inelastic rates, yl(xe) and
    a link centric decomposition ([17]) and implicitly
                                                                  zl(xi); with yl and zl for brevity. Thus, the
    assumes that packets injected into the source
                                                                  optimization problem is to maximize the sum of
    nodes by the flows, denoted by x[t], arrive at the
                                                                  utilities of elastic flows when guaranteeing that
    downstream nodes instantaneously. In reality,
                                                                  inelastic flows are supported. It is not difficult to
    packets will reach downstream nodes only after a
                                                                  show that the optimum value of FNO is an upper
    queueing and propagation delay incurred in the
                                                                  bound for the optimum value of SNO. To see
    intermediate nodes. It is shown in prior works
                                                                  this, note that any solution {x(t)}t≥0 that solve
    ([27], [5], [28], [17]) that the inclusion of these
    dynamics do not affect the long-term stability and            SNO must also satisfy                         where
    fairness characteristics of the system, and can be
    added to our queueing architecture by introducing
    a regulator queue before the queues associated                and is defined similarly. Otherwise, the queue l
    with each link. Thus, in this work we use the                 cannot be stable. This is equivalent to condition
                                                                  (4) in FNO. Thus, FNO contains all the feasible




August Issue                                          Page 50 of 68                                      ISSN 2229 5208
                                                          International Journal of Computer Information Systems,
                                                                                               Vol. 5, No.2, 2012

    points of SNO. In Section III-B, we will design
    an algorithm under which SNO can get arbitrarily
    close to the FNO solution, and thus guarantees
    the optimality of SNO. We start by showing that             This lemma implies that considering an inelastic
    there exists a unique Xe={xe}feЄFe that solves              flow fi,all routes in the optimal solution with a
    the FNO problem under Assumption 2 and 3 1.                 positive flow have the same value of ß. We note
    Proposition 1: If Assumption 2 and 3 hold, then             that αl of FNO is closely associated with the
    the X*e={xe}feЄFe that solves the network                   queue length pl of SNO, and correspondingly βR
    optimization problem is unique.                             of FNO is closely associated with the aggregate
    Proof: The optimization problem has a unique                queue length on a route qR of SNO. Such
    solution because the utility functions are strictly         connections are revealed and exploited in several
    concave, and constraints (4) and (5) are linear. To         earlier works for designing different network
    solve the FNO problem, we construct a partial               algorithms (e.g. [15], [16], [6], [7], [26]). Joint
    Lagrangian. Define αl to be the Lagrange                    Congestion Control and Load Balancing
    multipliers associated with constraint (4). Then,           Algorithm for the FNO problem:
    the partial Lagrangian can be written as                    1.Queue evolution for link l:


                                                                                                             ,
                                                                Where (v(t))+p(t) is zero if v(T)<0and P(t)=0;and
                                                                v(t) otherwise.
                                                                Congestion Controller for elastic flow fe:
    and
    Since FNO problem satisfies Slater’s condition                                             Load        Balancing
    ([4]) due to Assumption 2, the strong duality               implemented for inelastic flow fi:
    holds. We can then conclude that there exists x*
    :=(x*e,x*i),and α*:=(αl)l
    1.x* solves FNO problem; and 2.x*Є arg                                                           (8)
    maxx≥0 L(xi,xe,α*).Note that
                                                                and Where          satisfies


                                                  -
                                                                                                     (9)

    where βR := ΣlЄLR[l] αl. This decomposition
                                                                 and
    suggests that (i) The elastic flow fe should
    allocate its rates such that
                                                                when the system reaches the equilibrium, we
                                                                have xi(r)(t) = 0 for all r. This implies that
                                                                qRt(r)(t)= (t) fpr xi(r)(t)>0 and qRt(r)(t)>=
                                                (6)             for xi(r)(t)=0 . Thus, at the equilibrium point,
    (ii) The inelastic flow fi, should distribute its           qRt(r)(t) satisfies Lemma 1. Furthermore, from
                                                                (9), it is easy to see that
    packets over its available routes
    such that

                                                                                      for all t      (10)
                                                  (7)
                                                                Next, we will show the stability and optimality of
                                                                our joint congestion control and load balancing
    Since the optimization problem (7) has a linear             algorithm.
    objective,the following lemma holds ([4]):                  Proposition 2: Under Assumption 1, 2 and 3, the
    Lemma 1: For any R(r)iЄRi, we have:                         joint congestion control and load balancing
                                                                algorithm is globally asymptotically stable, i.e.
                                                and             limt→∞xe(t) = x*e starting from any x(0); where




August Issue                                          Page 51 of 68                                   ISSN 2229 5208
                                                           International Journal of Computer Information Systems,
                                                                                                Vol. 5, No.2, 2012

    x*e is the optimal solution to the FNO problem.              congestion controller is inclined to inject more
    Furthermore, (10) holds.                                     packets into the network with larger K. Also note
    B. Stochastic Model                                          that the load balancing implementation is slightly
    We now return to the original SNO problem with               different from the fluid model version to
    a minor variation: SNO Problem with Parameter                accommodate the randomness in the arrival
                                                                 processes for inelastic flows. The update is
                                                                 modified           to         ensure          that
    K (SNO-K):
    The Queue Evolution as in(1); Network Stability                                          holds for all t: The next
    as in(2);                                                    proposition establishes the stability and
                                                                 optimality of the joint algorithm for the stochastic
                                                                 system. Proposition 3: Under Assumptions 1, 2, 3
                                                                 and 4, the joint congestion control and load
                                                                 balancing algorithm stabilizes the system in the
    where K is a positive design parameter. We will
                                                                 sense that the Markov chain (p[t],xi[t]) is positive
    see that K parameter is critical in eliminating the
                                                                 recurrent with
    effect of randomness in the stochastic system on
    the long-term performance. Note that the solution
    to the SNO-K problem is independent of the
    value of K, and its optimum solution is identical
    to the solution of the SNO problem. Motivated by                                                ,
    the analysis in the fluid model, we propose the              and guarantees that the rate allocation satisfies,
    following joint congestion control and load
    balancing algorithm: Joint Congestion Control
    and Load Balancing Algorithm for the SNO-K
    problem:
    1.    Queue evolution for            a link       l:
                                                                 Here x*e is the optimal solution to the SNO-K
                                                                 problem,       is an arbitrarily chosen positive
    Congestion Controller for elastic flow fe:                   constant, and Є and ~c3 are positive values.
                                                                 IV. EXTENSION OF THE ALGORITHM
                                                                 A. Virtual Queue Algorithm
    where M is a positive constant satisfies                     Inelastic applications are delay sensitive, hence
                                                                 we assume that packets from inelastic flows have
                                                                 strict priority over their elastic counterparts.
    2.Load Balancing implemented for inelastic flow              Thus, the inelastic flows do not see the elastic
    fi:                                                          flows in the queues they traverse. But in some
                                                                 cases a link might be critically loaded by the
                                                                 inelastic traffic itself, thus resulting in large
                                                                 delays. Also, elastic traffic may have some delay
                                                   or            constraints that are non-negligible. An effective
    equivalently,                                                way of reducing the experienced delay is by
                                                                 including virtual queues that are served at a
                                                                 fraction of the actual service rate, and by using
                                                                 the virtual queue-length values as prices ([12]).
    Where           satisfies                                    To that end, we introduce two types of virtual
                                                                 queues with parameters, ρ1 and ρ2, which control
                                                                 the total load and the inelastic flow load,
                                                    and          respectively. Here for simplicity, we go back to
                                                                 the fluid model to design and analyze the joint
                                                                 congestion control and load balancing algorithm
                                                                 using virtual queues. We would like to have

    Remark: The factor 1=K in the congestion control             ~x*=         solve the following optimization
    equation comes from the factor K in the                      problem.
    optimization problem. It can be interpreted as the           FNO Problem with Virtual Queues (FNO-VQ):
    aggressiveness factor of the elastic flow, as the




August Issue                                         Page 52 of 68                                       ISSN 2229 5208
                                                                   International Journal of Computer Information Systems,
                                                                                                        Vol. 5, No.2, 2012



                                          (11)               and                                                         and




                                                                         Remark: In the above algorithm, note that the
                                                                         congestion control algorithm only responds to the
    Where 0< ρ2<= ρ1<1 To guarantee the feasibility                      virtual queues for elastic flows, but the load
    of the optimization problem above, we replace                        balancing algorithm responds to both the virtual
    our earlier Assumption 2 with: Assumption 5:                         queues for elastic flow and inelastic flows.
    There exists a xi such that                                          Further, the actual queue length is not used in the
                                                                         algorithm. In the following proposition, we show
                                                                         the equilibrium point of the algorithm with virtual
                                                                         queue is the optimal solution of (11).
                                                             and         Proposition 4: Under Assumptions 1, 3 and 5, the
                                                                         joint congestion control and load balancing
                                                                         algorithm for the FNO-VQ problem is globally

    To solve the FNO-VQ problem, we first                                asymptotically stable, i.e.,                  =
    introduce virtual queues for elastic and inelastic                   where          is the solution of the network
    flows on each link respectively. The virtual queue                   optimization
    length µl(t) for elastic flows evolutes as follows:                  Thus, we focus on the dynamics of file arrivals
                                                                         and departures, where we assume that
                                                                         (i)The number of inelastic flows are fixed.
    The virtual queue length for inelastic flows                         (ii) Files belonging to elastic flows dynamically
    evolves                as                 follows:                   arrive and depart. We assume that files arrive
                                                                         according to independent Poisson processes with
                                                                         rate     files/sec, the file sizes are independently
    Note that when the total instantaneous traffic load                  and exponentially distributed, with mean file size
    is larger than ρ1cl or the inelastic traffic load is                        bits. Further, let ne(t) denote the number of
    larger than ρ2cl, the virtual queues will build up,                  files belonging to flow fe in the network at time t.
    and the network controller will reduce the traffic                   The following result states that our joint
    load. Based on this virtual-queue scheme, we                         congestion control and load-balancing algorithm
    have the following joint congestion control and                      maximizes the network throughput region.
    load balancing algorithm:Joint Congestion                            Proposition 5: Under Assumptions 1-3, the
    Control and Load Balancing Algorithm for FNO-                        network is stable under the joint congestion
    VQ problem:Virtual queue evolution for a link l:                     control and load balancing algorithm if there
    Elastic
                                                                         exists   such that
     flows:                                              ;
    Inelasticflows:
                                                                                                   and        for        any
                                                 .

    Congestion        Controller   for   elastic         flow

                                                                                                                     . (12)
                                    whereis the
    aggregated virtual queue length of the elastic                       Under Assumptions 1, 4, and 5, the network is
    flow.     1. Load Balancing implemented for                          stable under the joint congestion control and load
    inelastic flow fi:                                                   balancing algorithm with virtual queues if there
                                                                         exists    such that
                                                     ,
    Where                                          satisfies



August Issue                                                 Page 53 of 68                                      ISSN 2229 5208
                                                            International Journal of Computer Information Systems,
                                                                                                 Vol. 5, No.2, 2012

                                                                  on the bottleneck link (2; 5). As a reaction to the
                                                                  increased contention from the elastic flow, the
                                                                  load balancing mechanism of the inelastic flows
                       And for any                   we           automatically pushes more and more traffic of the
                                                                  inelastic flow onto route 1.Note that although
                                                                  increasing the aggressiveness of the elastic flow
                                                                  will increase the utilization of the network, it will
    have                                                          result in more delays on the network flows as the
                                                                  queue length over the whole network grows, as
                                                                  shown by Proposition 3. Proposition 3 also
                                                                  suggests that larger K resulting better
                                          (13)                    convergence to the optimal operating point,
                                                                  which is confirmed in the above simulation.
    V. SIMULATION RESULTS
    In this section, we provide the simulation results            B. More Complex Topology To illustrate other
    for our algorithms under the stochastic model                 facets of our algorithm, we conducted our
    where the arrival process of the inelastic flow fi is         simulation in a more complicated network with
    such that Ai[t] is Poisson distributed with mean ai           different flow assignments. The topology of the
    for each t: A.The Effect of the Aggressiveness of             network is shown in Figure 5. The capacity of all
    the Inelastic Flow: We noted in Section III-B that            the links in the network is 20, and we used elastic
    the factor K represents the ‘aggressiveness’ of the           flows with identical utility functions and K = 200
    elastic flows. Also, it is revealed in Proposition 3          in our simulation, expecting close to optimal
    that K can be used to control the proximity to the            utilization (as shown in Figure 4). We simulate a
    optimal allocation. Here, we test these results for           sequence of scenarios discussed in five phases. In
    the case of proportionally fair allocation, which             Phase 1, two disjoint inelastic flows with the
    corresponds to having the utility function is                 routes as shown in Figure 5 share the network,
    chosen as ([25]): Ue(x) = αln x, thus                         having rates ai1 = 20 and ai2 = 10. The average
                                                                  rates provided on each route by our joint
                                                                  algorithm are given in Figure 5. When the two
                                                                  inelastic flows share a common bottleneck link in
                         and In this first set of
                                                                  Phase 2, the load balancing algorithm will shift
    simulations, we considered the network shown in
                                                                  part of the traffic from the bottleneck link to yield
    Figure 3 with the indicated link capacities and
                                                                  the average rates given in Figure 6.
    inelastic and elastic flows. Note that the arrival
    rate of the inelastic flow is ai = 15 to be
                                    distributed over




                                                                  Fig. 5. Phase 1: Two inelastic flows Fig.6.
                                                                  Phase 2: Two inelastic flows
                                                                  with disjoint routes. with intersecting routes.
                                                                  In Phase 3, an elastic flow enters the system and
    the two dashed routes.The joint                               shares a link with fi2 as in Figure 7. We can see
    fig 3:topology of network                                     from the average rates given in the figure that this
    fig 4:effect of k on each route                               elastic flow not only has an effect on fi2 but also
    algorithm for the SNO-K problem is                            shifts the rate of fi1. Here, it can be seen that the
    implemented for this network and the mean                     interaction between the flows becomes complex
    elastic rate allocation is computed for different             even for small networks, and it is not clear what
    values of K: Figure 4 illustrates the effect` . of K          the best allocation is. Yet, through our joint
    on the rates of the elastic flow and the                      algorithm, fe1 is able to operate dynamically
    distribution of the inelastic flow’s rate over its            close to the full capacity of all the resources
    available routes. We see that as the elastic flow             available to it. After adding another elastic flow
    becomes more aggressive, it achieves a higher                 fe2 into the network which is disjoint with all
    throughput and thus consumes greater resource                 other flows in Phase 4 shown in Figure 8, we can




August Issue                                          Page 54 of 68                                      ISSN 2229 5208
                                                            International Journal of Computer Information Systems,
                                                                                                 Vol. 5, No.2, 2012

    see that it has no effect on the rates of all other           implementation on delay. The simulation is
    routes, and it fully utilizes that route.                     conducted in the network showing in Figure 11.
                                                                  The parameter ρ1was set to 0.95 and ρ2 was set
                                                                  to 0.9 over all links. Table 1




    Fig. 7. Phase 3: Two intersecting inelastic flows,
    fig8:phase 4:2 intersecting inelastic
    and one elastic flow that interacts with
    them.Flows and 2 with disjoint routes. In Phase 5,
    a third elastic flow fe3 enters and shares common
    links with both fi1 and fi2, as shown in Figure 9.
    We can see that since fe1 also shares links with
    fi2, fe3 also has effect on it. It can be easily
    verified that xe1 = xe3 = 15 is the optimal
    operating point, and the average rates achieved by
    our algorithm is very close to optimal as
    predicted by Proposition 3. To study the
    importance of dynamic load balancing, we also
                                                                  (11)
    simulated a static rate distribution algorithm as a
    basis for comparison. This algorithm equally
    splits the inelastic traffic onto each of its routes
                                                                  Fig.      11.         Network         topology
    (assume it is feasible in the network), and does
                                                                  TABLE I:rate and delay on each routes
    the congestion control of the elastic flows in the
    same manner as in our algorithm. This algorithm
                                                                  RESULT & ANALYSIS:
    is implemented for the scenario in Phase 5 with
    the average rates indicated in Figure 10. We see
    that due to the absence of dynamic load
    balancing, the elastic flows cannot utilize the
    network fully since the rates assigned to the
    inelastic flows are fixed.Under the logarithm
    utility function, this approach achieves a utility of
    1:61αK while our algorithm achieves 2:71αK on
    the elastic flow fe3.




    Fig. 9. Phase 5: A third elastic flow enters
    Fig. 10. Performance under static that intersects
    with two inelastic flows. rate distribution for
    inelastic traffic.
    C. Simulation using the Virtual Queue Algorithm
    In this simulation, we use the joint congestion               Fig:12 throughput of network with different
    control and load balancing algorithm with virtual             protocols
    queue to show the impact of the virtual queue




August Issue                                          Page 55 of 68                                    ISSN 2229 5208
                                                          International Journal of Computer Information Systems,
                                                                                               Vol. 5, No.2, 2012

    Table I and fig12 compares the performance
    with and without virtual queues. As we can                  REFERENCES
    observe, under the original algorithm, Route 1 is           [1] E. Altman, T. Bas¸ar, and R. Srikant. Congestion control
                                                                as a stochastic control problem with action delays.
    critically loaded by inelastic flow, resulting in a         Automatica, pages 1937–1950, 1999. Special Issue on
    large delay. With the virtual queue                         Control Methods for communication networks, V.
    implementation, we manage to decrease the rate              Anantharam and J. Walrand, editors.
    on Route 1, thus dropping the delay significantly.          [2] J. Bolot and A. Shankar. Dynamic behavior of rate-based
                                                                flow control mechanisms. ACM Comput. Commun. Rev.,
    As one can observe from the table, the delay is             20(2), 1992.
    greatly reduced for both elastic and inelastic              [3] S. Borst and N. Hegde. Integration of streaming and
    traffic without a significant degradation in the            elastic traffic in wireless networks. In Proc. INFOCOM, May
    rate of the elastic traffic.Thus, especially under          2007.
                                                                [4] S. Boyd and L. Vandenberghe. Convex Optimization.
    critical loaded scenarios, virtual queue                    Cambridge University Press, Cambridge, 2004.
    implementation can achieve significant delay                [5] L. Bui, A. Eryilmaz, R. Srikant, and X. Wu. Joint
    improvements.                                               asynchronous congestion control and distributed scheduling
    VI. CONCLUSION                                              for wireless networks. Proceedings of IEEE Infocom 2006.
                                                                [6] A. Eryilmaz and R. Srikant. Fair resource allocation in
    In this work, we consider the optimal control of            wireless networks using queue-length based scheduling and
    networks that serve heterogeneous traffic types             congestion control. In Proceedings of IEEE Infocom, volume
    with diverse demands, namely inelastic and                  3, pages 1794–1803, Miami, FL, March 2005.
    elastic traffic. We formulated a new network                [7] A. Eryilmaz and R. Srikant. Resource allocation of multi-
                                                                hop wireless networks. In Proceedings of International
    optimization problem and proposed a novel                   Zurich Seminar on Communications, February 2006.
    queueing architecture, and develop a distributed            [8] L. Georgiadis, M. J. Neely, and L. Tassiulas. Resource
    load balancing and congestion control algorithm             Allocation and Cross-Layer Control in Wireless Networks.
    with provably optimal performance. We also                  2006. Foundations and Trends in Networking.
                                                                [9] P. Hande, S. Zhang, and M. Chiang. Distributed rate
    provided an important improvement to our joint              allocation for inelastic flows. Networking, IEEE/ACM
    algorithm to achieve better delay performance by            Transactions on, 15(6):1240– 1253, Dec. 2007.
    introducing new design parameters (½1; ½2)                  [10] K. Kar, S. Sarkar, and L. Tassiulas. A simple rate
    together with a set of virtual queues. We have              control algorithm for maximizing total user utillity. In
                                                                Proceedings of IEEE INFOCOM, Anchorage, Alaska, 2001.
    also extended our algorithm to the case of                  [11] F. P. Kelly, A. Maulloo, and D. Tan. Rate control in
    dynamic arrivals and departures of the flows.               communication networks: Shadow prices, proportional
    Such a scenario is relevant to real-world                   fairness and stability. Journal of the Operational Research
    operation as the real-world applications randomly           Society, 49:237–252, 1998.
                                                                [12] S. Kunniyur and R. Srikant. Analysis and design of an
    initiate flows that lasts for a random duration.            adaptive virtual queue algorithm for active queue
    Future research of this topic includes: (i) One             management. IEEE/ACM Transactions on Networking, pages
    future direction is to extend our results to multi-         286–299, April 2004. An earlier version appeared in Proc.
    hop wireless networks with fading channels and              ACM Sigcomm 2001.
                                                                [13] J. Lee, R. R. Mazumdar, and N. Shroff. Non-convex
    interference,    and     develop      joint   load-         optimization and rate control for multi-class services in the
    balancing/congestion control/routing/scheduling             internet. IEEE/ACM Trans. on Networking, 13(4):827 – 840,
    algorithms. (ii) Here, we considered a time-                Aug 2005.
    slotted system, and assumed that the network is             [14] R. Li, L. Ying, A. Eryilmaz, and N. B. Shroff. A unified
                                                                approach to optimizing performance in networks serving
    perfectly synchronized. The impact of possible              heterogeneous flows, 2008. Technical Report, available
    asynchronism on the algorithm performance                   online
    needs to be studied. (iii) We adopted a link-               ttp://www.ece.osu.edu/»eryilmaz/ElasticInelasticReport.pdf.
    centric formulation, which assumes instantaneous            [15] X. Lin and N. Shroff. Joint rate control and scheduling
                                                                in multihop wireless networks. In Proceedings of IEEE
    arrivals of the packets at all the links on their           Conference on Decision and Control, Paradise Island,
    routes. An alternative is to consider a nodecentric         Bahamas, December 2004.
    formulation, where packets are sequentially                 [16] X. Lin and N. Shroff. The impact of imperfect
    transferred, and a source only requires the                 scheduling on crosslayer rate control in multihop wireless
                                                                networks. In Proceedings of IEEE Infocom, Miami, FL,
    information of the queues at the source. (iv) Here          March 2005.
    we restricted each elastic flow to a single route.          [17] X. Lin, N. B. Shroff, and R. Srikant. A tutorial on cross-
    An extension to multiple routes or dynamic                  layer optimization in wireless networks. IEEE Journal on
    routing for elastic flow constitutes a part of our          Selected Areas in Communications, special issue on
                                                                Nonlinear Optimization of Communication Systems, 14(8),
    ongoing work. (v) So far, we have focused on the            Aug 2006.
    stability and long term guarantees for the traffic          [18] X. Lin, N. B. Shroff, and R. Srikant. On the connection-
    types. We aim to investigate oscillatory behavior           level stability of congestion-controlled communication
    ([20]) and delay characteristics in our future              networks. IEEE Transactions on Information Theory, 2008.
                                                                To appear.
    work.




August Issue                                        Page 56 of 68                                              ISSN 2229 5208
                                                                       International Journal of Computer Information Systems,
                                                                                                            Vol. 5, No.2, 2012

    [19] S. H. Low and D. E. Lapsley. Optimization flow control,
    I: Basic algorithm and convergence. IEEE/ACM
    Transactions on Networking, 7:861–875, December 1999.
    [20] E. Mallada and F. Paganini. Stability of node-based
    multipath routing and dual congestion control. Decision and
    Control, 2008. CDC 2008. 47th IEEE Conference on, pages
    1398–1403, Dec. 2008.
    [21] M. Neely, E. Modiano, and C. Li. Fairness and optimal
    stochastic control for heterogeneous networks. In
    Proceedings of IEEE Infocom, pages 1723–1734, Miami, FL,
    March 2005.
    [22] S. Patil and G. Veciana. Managing resources and quality
    of service in heterogeneous wireless systems exploiting
    opportunism. IEEE/ACM Trans. Netw., 15(5):1046–1058,
    2007.
    [23] D. Qiu and N. B. Shroff. A predictive flow control
    scheme for efficient network utilization and QoS.
    IEEE/ACM Transactions on Networking, 12(1), February
    2004.
    [24] S. Shakkottai and A. Stolyar. Scheduling algorithms for
    a mixture of real-time and non-real-time data in hdr. In in
    Proceedings of 17th International Teletraffic Congress (ITC-
    17, pages 793–804, 2001.
    [25] R. Srikant. The Mathematics of Internet Congestion
    Control. Birkh¨auser, Boston, MA, 2004.
    [26] A. Stolyar. Maximizing queueing network utility subject
    to stability: Greedy primal-dual algorithm. Queueing
    Systems, 50(4):401–457, 2005.
    [27] X. Wu and R. Srikant. Regulated maximal matching: A
    distributed scheduling algorithm for multi-hop wireless
    networks with nodeexclusive spectrum sharing. In
    Proceedings of IEEE Conference on
    Decision and Control., 2005.
    [28] L. Ying, R. Srikant, A. Eryilmaz, and G. E. Dullerud.
    Distributed fair resource allocation in cellular networks in the
    presence of heterogeneous delays. In Proceedings of
    International Symposium on Modeling and Optimization in
    Mobile, Ad Hoc, and Wireless Networks (WiOPT), 2005.


    AUTHORS PROFILE

                           1].A.RARAJU,Received        B.Tech
                           degree from Sri Venkateshwara
                           University,Trupathi.and M.Tech in
                           Computer Science & Engineering
                           from Kottam college of engineering,
                           Kurnool,JNTU Ananthapur, Andhra
                           Pradesh, India. raraju22@gmail.com




                        2]. K.SRINIVAS, received MCA
                        degree from Osmania University and
                        M.Tech in Computer Science &
                        Engineering from JNTU Hyderabad,
                        and Ph.D from Rayalaseema
                        University(submitted),    Andhra
                        Pradesh, India. And also CISCO
                        CCNA Certified. Currently he is
                        Professor and HOD Of Computer
                        Science Department in Kottam
                        College of Engineering, Kurnool,
    AP, India. kipgs2008@gmail.com




August Issue                                                    Page 57 of 68                                     ISSN 2229 5208

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:36
posted:9/16/2012
language:English
pages:10