Backpressure routing explained

In queueing theory, a discipline within the mathematical theory of probability, the backpressure routing algorithm is a method for directing traffic around a queueing network that achieves maximum network throughput, which is established using concepts of Lyapunov drift. Backpressure routing considers the situation where each job can visit multiple service nodes in the network. It is an extension of max-weight scheduling where each job visits only a single service node.

Introduction

Backpressure routing is an algorithm for dynamically routing traffic over a multi-hop network by using congestion gradients. The algorithm can be applied to wireless communication networks, including sensor networks, mobile ad hoc networks (MANETS), and heterogeneous networks with wireless and wireline components.[1] [2]

Backpressure principles can also be applied to other areas, such as to the study ofproduct assembly systems and processing networks.[3] This article focuses on communication networks,where packets from multiple data streams arrive andmust be delivered to appropriate destinations. The backpressurealgorithm operates in slotted time. Every time slot it seeks to route data in directions thatmaximize the differential backlog between neighboring nodes. This is similar to how waterflows through a network of pipes via pressure gradients. However, the backpressure algorithmcan be applied to multi-commodity networks (where different packets may have different destinations),and to networks where transmission rates can be selectedfrom a set of (possibly time-varying) options. Attractive featuresof the backpressure algorithm are: (i) it leads to maximum network throughput, (ii)it is provably robust to time-varying network conditions, (iii) itcan be implemented without knowing traffic arrival rates or channel stateprobabilities. However, the algorithm may introduce large delays, and maybe difficult to implement exactly in networks with interference. Modifications ofbackpressure that reduce delay and simplify implementation are described belowunder Improving delay and Distributed backpressure.

Backpressure routing has mainly been studied in a theoreticalcontext. In practice, ad hoc wireless networks have typicallyimplemented alternative routing methods based on shortestpath computations or network flooding, such asAd Hoc on-Demand Distance Vector Routing (AODV),geographic routing, and extremely opportunistic routing (ExOR).However, the mathematical optimality properties of backpressurehave motivated recent experimental demonstrations of its useon wireless testbeds at the University of Southern Californiaand at North Carolina State University.[4] [5]

Origins

The original backpressure algorithm was developed by Tassiulas and Ephremides.[1] They considered a multi-hop packet radio network with random packet arrivals and a fixed set of link selection options. Their algorithm consisted of a max-weight link selection stage and a differential backlog routing stage.An algorithm related to backpressure, designed for computing multi-commodity network flows, was developed in Awerbuch and Leighton.[6] The backpressure algorithm was later extended by Neely, Modiano, and Rohrs to treat scheduling for mobile networks.[7] Backpressure is mathematically analyzed via the theory of Lyapunov drift, and can be used jointly with flow control mechanisms to provide network utility maximization.[8] [9] [2] [10] [11] (see also Backpressure with utility optimization and penalty minimization).

How it works

Backpressure routing is designed to make decisions that (roughly) minimize the sum of squares of queue backlogs in the network from one timeslot to the next. The precise mathematical development of this technique is described inlater sections. This section describes the general network model and the operation of backpressure routing with respectto this model.

The multi-hop queueing network model

Consider a multi-hop network with N nodes (see Fig. 1 for an example with N = 6).The network operates inslotted time

t\in\{0,1,2,\ldots\}

. On each slot, new data can arrive tothe network, and routing and transmission scheduling decisions are madein an effort to deliver all data to its proper destination. Let data that is destinedfor node

c\in\{1,...,N\}

be labeled as commodity c data. Data in each node is stored according to its commodity. For

n\in\{1,\ldots,N\}

and

c\in\{1,\ldots,N\}

, let
(c)
Q
n

(t)

be the current amount of commodity c data in node n, also called the queue backlog. A closeup of the queue backlogs inside a node is shown in Fig. 2.The units of
(c)
Q
n

(t)

depend on the context of the problem.For example, backlog can take integer units of packets, which is useful in cases when data is segmented into fixed length packets. Alternatively, it can take real valued units of bits. It is assumed that
(c)
Q
c

(t)=0

for all

c\in\{1,\ldots,N\}

and all timeslots t, because no node stores data destined for itself. Every timeslot, nodes can transmit data to others. Data that is transmitted from one node to another node is removed from the queue of the first node and added to the queue of the second. Data that is transmitted to its destination is removed from the network. Data can also arrive exogenously to the network, and
(c)
A
n

(t)

is defined as the amount of new data that arrives to node n on slot t that must eventuallybe delivered to node c.

Let

\muab(t)

be the transmission rate used by the network over link (a,b) on slot t, representing the amount of data it can transfer from node a to node b on the current slot. Let

(\muab(t))

be the transmission rate matrix. These transmission rates must be selected within a set of possibly time-varying options. Specifically,the network may have time-varying channels and nodemobility, and this can affect its transmission capabilities every slot.To model this, let S(t) represent the topology state of the network, which capturesproperties of the network on slot t that affect transmission. Let

\GammaS(t)

represent the setof transmission rate matrix options available under topology state S(t).Every slot t, the network controller observes S(t) and chooses transmissionrates

(\muab(t))

within the set

\GammaS(t)

.The choice of which

(\muab(t))

matrixto select on each slot t is described in the next subsection.

This time-varying network model was first developed for the case when transmission rates every slot t were determined by general functions of a channel state matrix and a power allocation matrix.[7] The model can also be used when rates are determined by other control decisions, such as server allocation, sub-band selection, coding type, and so on. It assumes the supportable transmission rates are known and there are no transmission errors. Extended formulations of backpressure routing can be used for networks with probabilistic channel errors, including networks that exploit the wireless broadcast advantage via multi-receiver diversity.[12]

The backpressure control decisions

Every slot t the backpressure controller observes S(t) and performs the following 3 steps:

opt
c
ab

(t)

to use.

(\muab(t))

matrix in

\GammaS(t)

to use.
opt
c
ab

(t)

it will transmit over link (a,b) (being at most

\muab(t)

, but possibly being less in some cases).

Choosing the optimal commodity

Each node a observes its own queue backlogs and the backlogs in its currentneighbors. A current neighbor of node a is a node b such that it is possible to choosea non-zero transmission rate

\muab(t)

on the current slot.Thus, neighbors are determined by the set

\GammaS(t)

. In the extreme case, anode can have all N − 1 other nodes as neighbors. However, it is common to use sets

\GammaS(t)

that preclude transmissions between nodes that are separated by more than a certain geographic distance,or that would have a propagated signal strength below a certain threshold.Thus, it is typical for the number of neighborsto be much less than N − 1. The example in Fig. 1 illustrates neighbors by link connections, so that node 5 has neighbors 4 and 6. The example suggests a symmetric relationship between neighbors (so that if 5 is a neighbor of 4, then 4 is a neighbor of 5), but this need not be the case in general.

The set of neighbors of a given node determines the set of outgoing links it can use for transmission on the current slot. For each outgoing link (a,b), the optimal commodity

opt
c
ab

(t)

is defined as the commodity

c\in\{1,\ldots,N\}

that maximizes the following differential backlog quantity:
(c)
Q
a

(t)-

(c)
Q
b

(t)

Any ties in choosing the optimal commodity are broken arbitrarily.

An example is shown in Fig. 2. The example assumes each queue currently has only 3 commodities: red, green, andblue, and these are measured in integer units of packets. Focusing on the directed link (1,2), the differential backlogs are:

(red)
Q
1

(t)-

(red)
Q
2

(t)=1

(green)
Q
1

(t)-

(green)
Q
2

(t)=2

(blue)
Q
1

(t)-

(blue)
Q
2

(t)=-1

Hence, the optimal commodity to send over link (1,2) on slot t is the green commodity. On the other hand, the optimal commodity to send over the reverse link (2,1) on slot t is the blue commodity.

Choosing the μab(t) matrix

Once the optimal commodities have been determined for each link (a,b), the network controller computes the following weights

Wab(t)

:

Wab(t)=

opt(t))
(c
ab
max\left[Q
a

(t)-

opt
(c(t))
ab
Q
b

(t),0\right]

The weight

Wab(t)

is the value of the differential backlog associated with the optimal commodityfor link (a,b), maxed with 0. The controller then chooses transmission rates as the solution tothe following max-weight problem (breaking ties arbitrarily):

(Eq.1)    Maximize:

N\mu
\sum
ab

(t)Wab(t)

(Eq.2)    Subjectto:(\muab(t))\in\GammaS(t)

As an example of the max-weight decision, suppose that on the current slot t, the differential backlogs on each link of the 6 node network lead to link weights

Wab(t)

given by:

(Wab(t))=\left[\begin{array}{cccccc} 0&2&1&1&6&0\\ 1&0&1&2&5&6\\ 0&7&0&0&0&0\\ 1&0&1&0&0&0\\ 1&0&7&5&0&0\\ 0&0&0&0&5&0 \end{array} \right]

While the set

\GammaS(t)

might contain an uncountably infinite numberof possible transmission rate matrices, assume for simplicity that the current topology state admits only 4 possiblechoices:

\GammaS(t)=\{\boldsymbol{\mu}a,\boldsymbol{\mu}b,\boldsymbol{\mu}c,\boldsymbol{\mu}d\}

illustration of the 4 possible transmission rate selections under the current topology state S(t). Option (a) activatesthe single link (1,5) with a transmission rate of

\mu15=2

. All other options use two links, with transmission rates of 1 on each of the activated links.

These four possibilities are represented in matrix form by:

\boldsymbol{\mu}a=\left[\begin{array}{cccccc} 0&0&0&0&2&0\\ 0&0&0&0&0&0\\ 0&0&0&0&0&0\\ 0&0&0&0&0&0\\ 0&0&0&0&0&0\\ 0&0&0&0&0&0 \end{array} \right],\boldsymbol{\mu}b=\left[\begin{array}{cccccc} 0&0&0&0&0&0\\ 0&0&1&0&0&0\\ 0&0&0&0&0&0\\ 0&0&0&0&1&0\\ 0&0&0&0&0&0\\ 0&0&0&0&0&0 \end{array} \right]

\boldsymbol{\mu}c=\left[\begin{array}{cccccc} 0&0&0&0&0&0\\ 1&0&0&0&0&0\\ 0&0&0&0&0&0\\ 0&0&0&0&1&0\\ 0&0&0&0&0&0\\ 0&0&0&0&0&0 \end{array} \right],\boldsymbol{\mu}d=\left[\begin{array}{cccccc} 0&0&0&0&0&0\\ 0&0&0&0&0&0\\ 0&1&0&0&0&0\\ 0&0&0&0&0&0\\ 0&0&0&1&0&0\\ 0&0&0&0&0&0 \end{array} \right]

Observe that node 6 can neither send nor receive under any of these possibilities.This might arise because node 6 is currently out of communication range.The weighted sum of rates for each of the 4 possibilities are:

\sumabWab(t)\muab(t)=12

.

\sumabWab(t)\muab(t)=1

.

\sumabWab(t)\muab(t)=1

.

\sumabWab(t)\muab(t)=12

.

Because there is a tie for the maximum weight of 12, the network controller can break the tie arbitrarily bychoosing either option

\boldsymbol{\mu}a

or option

\boldsymbol{\mu}d

.

Finalizing the routing variables

Suppose now that the optimal commodities

opt
c
ab

(t)

have been determined for each link, and the transmissionrates

(\muab(t))

have also been determined.If the differential backlog for the optimal commodity on a given link (a,b) is negative, then no data is transferredover this link on the current slot. Else, the network offers to send

\muab(t)

units of commodity
opt(t)
c
ab
data over this link. This is done by defining routing variables
(c)
\mu
ab

(t)

for each link (a,b) andeach commodity c, where:
(c)
\mu
ab

(t)=\begin{cases} \muab(t)&ifc=

opt
c
ab

(t)and

opt
(c(t))
ab
Q
a
opt
(c(t))
ab
(t)-Q
b

(t)\geq0\\ 0&otherwise \end{cases}

The value of

(c)
\mu
ab

(t)

represents the transmission rate offered to commodity c data over link(a,b) on slot t.However, nodes might not have enough of a certain commodity to support transmissionat the offered rates on all of their outgoing links. This arises on slot t for node n and commodity c if:
(c)
Q
n

(t)<

(c)
\sum
nb

(t)

In this case, all of the

(c)
Q
n

(t)

data is sent, and null data is used to fill the unused portions of the offered rates,allocating the actual data and null data arbitrarily over the corresponding outgoing links (according to the offered rates).This is called a queue underflow situation. Such underflows do not affect the throughputor stability properties of the network. Intuitively, this is because underflowsonly arise when the transmitting node has a low amount of backlog, which means thenode is not in danger of instability.

Improving delay

The backpressure algorithm does not use any pre-specified paths. Paths are learneddynamically, and may be different for different packets. Delay can be very large, particularly when the system is lightlyloaded so that there is not enough pressure to push data towards the destination. As an example, suppose one packetenters the network, and nothing else ever enters. This packet may take a loopy walk through the network and never arriveat its destination because no pressure gradients build up. This does not contradict the throughput optimality or stabilityproperties of backpressure because the network has at most one packet at any time and hence is trivially stable (achieving a delivery rate of 0, equal to the arrival rate).

It is also possible to implement backpressure on a set of pre-specified paths. This can restrict the capacity region, but might improve in-orderdelivery and delay. Another way to improve delay, without affecting the capacity region, is to use an enhancedversion that biases link weights towards desirable directions.[7] Simulations of such biasing have shown significant delay improvements.[12] [2] Note that backpressure does not require First-in-First-Out (FIFO) service at the queues. It has been observedthat Last-in-First-Out (LIFO) service can dramatically improve delay for the vast majority of packets,without affecting throughput.[13] [14]

Distributed backpressure

Note that once the transmission rates

(\muab(t))

have been selected, the routing decision variables
(c)
\mu
ab

(t)

can be computed in a simple distributed manner, where each node only requires knowledge ofqueue backlog differentials between itself and its neighbors. However, selection of the transmission rates requires a solution to themax-weight problem in Eqs. (1)-(2). In the special case when channels are orthogonal, the algorithm has a natural distributed implementation and reduces to separate decisions at each node. However, the max-weight problem is a centralized control problem for networks with inter-channel interference. It can also be very difficult to solve even in a centralized way.

A distributed approach for interference networks with link rates that are determined by the signal-to-noise-plus-interference ratio (SINR) can be carried out using randomization.[7] Each node randomly decides to transmit every slot t (transmitting a "null" packet if it currently does nothave a packet to send). The actual transmission rates, and the corresponding actual packets to send,are determined by a 2-step handshake:On the first step, the randomly selected transmitter nodes send a pilot signal with signal strength proportionalto that of an actual transmission. On the second step,all potential receiver nodes measure the resulting interference and send that information back to thetransmitters. The SINR levels for all outgoing links (n,b) are then known to all nodes n,and each node n can decideits

\munb(t)

and
(c)
(\mu
nb

(t))

variables based on this information.The resulting throughput is not necessarily optimal. However, the random transmission process can be viewed as a part of the channel state process (provided that null packets are sent in cases of underflow, so that the channel state process does not depend on past decisions). Hence, the resulting throughput of this distributed implementation is optimal over the class of all routing and scheduling algorithms that use such randomized transmissions.

Alternative distributed implementations can roughly be grouped into two classes:The first class of algorithms consider constant multiplicative factor approximations to the max-weight problem,and yield constant-factor throughput results. The second class of algorithms consider additive approximations to the max-weightproblem, based on updating solutions to the max-weight problem over time. Algorithms in this second class seem to require static channelconditions and longer (often non-polynomial) convergence times, although they can provably achieve maximum throughputunder appropriate assumptions.[15] [3] [11] Additive approximations are often usefulfor proving optimality of backpressure when implemented with out-of-date queue backlog information (see Exercise 4.10 of the Neely text).[11]

Mathematical construction via Lyapunov drift

This section shows how the backpressure algorithm arises as a natural consequence ofgreedily minimizing a bound on the change in the sum of squares of queue backlogs from one slot to the next.[7] [2]

Control decision constraints and the queue update equation

Consider a multi-hop network with N nodes, as described in the above section.Every slot t, the network controller observes the topology state S(t) and choosestransmission rates

(\muab(t))

and routing variables
(c)
(\mu
ab

(t))

subjectto the following constraints:

(Eq.3)    (\muab(t))\in\GammaS(t)

(Eq.4)    0\leq

(c)
\mu
ab

(t)    \foralla,b,c,\forallt

(Eq.5)   

(c)
\sum
ab

(t)\leq\muab(t)    \forall(a,b),\forallt

Once these routing variables are determined, transmissions are made (using idle fill if necessary), and the resulting queuebacklogs satisfy the following:

(Eq.6)   

(c)
Q
n

(t+1)\leq

(c)
max\left[Q
n

(t)-

(c)
\sum
nb

(t),0\right]+

(c)
\sum
an

(t)+

(c)
A
n

(t)

where

(c)
A
n

(t)

is the random amount of new commodity cdata that exogenously arrives to node n on slot t, and
(c)
\mu
nb

(t)

is the transmission rate allocatedto commodity c traffic on link (n,b) on slot t. Note that
(c)
\mu
nb

(t)

may be more than the amount ofcommodity c data that is actually transmitted on link (a,b) on slot t. This is because there may not be enough backlogin node n. For this same reason, Eq. (6) is an inequality, rather than an equality, because
(c)
\sum
an

(t)

may be more than the actual endogenous arrivals of commodity c to node n on slot t.An important feature of Eq. (6) is that it holds even if the
(c)
\mu
ab

(t)

decision variables are chosen independently of queue backlogs.

It is assumed that

(c)
Q
c

(t)=0

for all slots t and all

c\in\{1,\ldots,N\}

, as no queue stores data destined for itself.

Lyapunov drift

Define

\boldsymbol{Q}(t)=

(c)
(Q
n

(t))

as the matrix of current queue backlogs.Define the following non-negative function, called a Lyapunov function:

L(t)=

1
2
N
\sum
c=1
(c)
Q
n

(t)2

This is a sum of the squares of queue backlogs (multiplied by 1/2 only for convenience in later analysis). The above sum is the same as summing over all n, c such that

nc

because
(c)
Q
c

(t)=0

for all

c\in\{1,\ldots,N\}

and all slots t.

The conditional Lyapunov drift

\Delta(t)

is defined:

\Delta(t)=E\left[L(t+1)-L(t)\mid\boldsymbol{Q}(t)\right]

Note that the following inequality holds for all

q\geq0

,

a\geq0

,

b\geq0

:

(max[q-b,0]+a)2\leqq2+b2+a2+2q(a-b)

By squaring the queue update equation (Eq. (6)) and using the above inequality, it is not difficultto show that for all slots t and under any algorithm for choosing transmission and routing variables

(\muab(t))

and
(c)
(\mu
ab

(t))

:[2]

(Eq.7)    \Delta(t)\leqB+

(c)
\sum
n
(c)
(t)E\left[λ
n

(t)+

(c)
\sum
an

(t)-

(c)
\sum
nb

(t)|\boldsymbol{Q}(t)\right]

where B is a finite constant that depends on the second moments of arrivals and the maximum possible second moments of transmission rates.

Minimizing the drift bound by switching the sums

The backpressure algorithm is designed to observe

\boldsymbol{Q}(t)

andS(t) every slot t and choose

(\muab(t))

and
(c)
(\mu
ab

(t))

to minimize the right-hand-side of the drift bound Eq. (7). Because B is a constant and
(c)
λ
n
are constants, this amounts to maximizing:
(c)
E\left[\sum
n

(t)\left[

(c)
\sum
nb

(t)-

(c)
\sum
an

(t)\right]|\boldsymbol{Q}(t)\right]

where the finite sums have been pushed through the expectations to illuminate the maximizing decision.By the principle of opportunistically maximizing an expectation, the above expectation is maximized bymaximizing the function inside of it (given the observed

\boldsymbol{Q}(t)

,

S(t)

).Thus, one chooses

(\muab(t))

and
(c)
(\mu
ab

(t))

subject to the constraints Eqs. (3)-(5) to maximize:
(c)
\sum
n

(t)\left[

(c)
\sum
nb

(t)-

(c)
\sum
an

(t)\right]

It is not immediately obvious what decisions maximize the above. This can be illuminated by switching the sums.Indeed, the above expression is the same as below:

(c)
\sum
ab
(c)
(t)[Q
a

(t)-

(c)
Q
b

(t)]

The weight

(c)
Q
a

(t)-

(c)
Q
b

(t)

is called the current differential backlog of commodity c betweennodes a and b. The idea is to choose decision variables
(c)
(\mu
ab

(t))

so as to maximize the aboveweighted sum, where weights are differential backlogs. Intuitively, this means allocating larger rates in directionsof larger differential backlog.

Clearly one should choose

(c)
\mu
ab

(t)=0

whenever
(c)
Q
a

(t)-

(c)
Q
b

(t)<0

. Further, given

\muab(t)

for a particular link

(a,b)

,it is not difficult to show that the optimal
(c)
\mu
ab

(t)

selections,subject to Eqs. (3)-(5),are determined as follows: First find the commodity
opt
c
ab

(t)\in\{1,\ldots,N\}

that maximizes the differential backlog for link (a,b).If the maximizing differential backlog is negative for link (a,b),assign
(c)
\mu
ab

(t)=0

for all commodities

c\in\{1,\ldots,N\}

on link (a,b). Else, allocate the full link rate

\muab(t)

to the commodity
opt
c
ab

(t)

, and zero rate to all other commodities on this link. With this choice, it follows that:
(c)
\sum
ab
(c)
(t)[Q
a

(t)-

(c)
Q
b

(t)]=\muab(t)Wab(t)

where

Wab(t)

is the differential backlog of the optimal commodity for link (a,b) on slot t (maxed with 0):

Wab(t)=max[

opt
(c(t))
ab
Q
a

(t)-

opt
(c(t))
ab
Q
b

(t),0]

It remains only to choose

(\muab(t))\in\GammaS(t)

. This is done by solving the following:

Maximize:

N\mu
\sum
ab

(t)Wab(t)

Subjectto:(\muab(t))\in\GammaS(t)

The above problem is identical to the max-weight problem in Eqs. (1)-(2).The backpressure algorithm uses the max-weight decisions for

(\muab(t))

, and then chooses routing variables
(c)
(\mu
ab

(t))

via the maximum differential backlog as described above.

A remarkable property of the backpressure algorithm is that it acts greedily every slot t based only on the observed topology state S(t) and queue backlogs

\boldsymbol{Q}(t)

for that slot. Thus, it does not require knowledge of the arrival rates
(c)
(λ
n

)

or the topology state probabilities

\piS=Pr[S(t)=S]

.

Performance analysis

This section proves throughput optimality of the backpressure algorithm.[2] [11] For simplicity, the scenario where events are independent and identicallydistributed (i.i.d.) over slots is considered, although the same algorithm can be shown to work in non-i.i.d. scenarios (seebelow under Non-i.i.d. operation and universal scheduling).

Dynamic arrivals

Let

(c)
(A
n

(t))

be the matrix of exogenous arrivals on slot t. Assume this matrix is independent and identicallydistributed (i.i.d.) over slots with finite second moments and with means:
(c)
λ
n

=

(c)
E\left[A
n

(t)\right]

It is assumed that

(c)
λ
c

=0

for all

c\in\{1,\ldots,N\}

, as no data arrives that is destined for itself. Thus,the matrix of arrival rates
(c)
(λ
n

)

is a

N x N

matrix of non-negative real numbers, with zeros on the diagonal.

Network capacity region

Assume the topology state S(t) is i.i.d. over slots with probabilities

\piS=Pr[S(t)=S]

(if S(t) takes values in an uncountably infinite set of vectors with real-valued entries,then

\piS

is a probability distribution, not a probability mass function).A general algorithm for the network observes S(t) every slot t and chooses transmission rates

(\muab(t))

and routing variables
(c)
(\mu
ab

(t))

according to the constraints in Eqs. (3)-(5). The network capacity region

Λ

is the closure of the set of all arrival rate matrices
(c)
(λ
n

)

for which there exists an algorithm that stabilizes the network. Stability of all queues implies that the total input rate of traffic into the network is the same as the total rate of data delivered to its destination. It can be shown that for any arrival rate matrix
(c)
(λ
n

)

in the capacity region

Λ

, there is a stationary and randomized algorithm that chooses decision variables
*(t))
(\mu
ab
and
*(c)
(\mu
ab

(t))

every slot t based only on S(t) (and hence independently of queue backlogs)that yields the following for all

nc

:[7] [11]

(Eq.8)   

(c)
E\left[λ
n

+

*(c)
\sum
an

(t)-

*(c)
\sum
nb

(t)\right]\leq0

Such a stationary and randomized algorithm that bases decisions only on S(t) is called an S-only algorithm. It is often useful to assume that

(c)
(λ
n

)

is interior to

Λ

, so that there is an

\epsilon>0

such that
(c)
(λ
n

+\epsilon

(c)
1
n

)\inΛ

, where
(c)
1
n
is 1 if

nc

, and zero else. In that case, there is an S-only algorithm that yields the following for all

nc

:

(Eq.9)   

(c)
E\left[λ
n

+

*(c)
\sum
an

(t)-

*(c)
\sum
nb

(t)\right]\leq-\epsilon

As a technical requirement, it is assumed that the second moments of transmission rates

\muab(t)

are finite under any algorithm for choosing these rates. This trivially holds if there is a finite maximum rate

\mumax

.

Comparing to S-only algorithms

Because the backpressure algorithm observes

\boldsymbol{Q}(t)

and S(t) every slot t and chooses decisions

(\muab(t))

and
(c)
(\mu
ab

(t))

to minimize the right-hand-side of the drift bound Eq. (7), we have:

(Eq.10)    \Delta(t)\leqB+

(c)
\sum
n
(c)
(t)E\left[λ
n

(t)+

*(c)
\sum
an

(t)-

*(c)
\sum
nb

(t)|\boldsymbol{Q}(t)\right]

where

*(t))
(\mu
ab
and
*(c)
(\mu
ab

(t))

are any alternative decisions that satisfy Eqs. (3)-(5), including randomized decisions.

Now assume

(c)
(λ
n

)\inΛ

. Then there exists an S-only algorithm that satisfiesEq. (8). Plugging this into the right-hand-side of Eq. (10) and noting that the conditional expectation given

\boldsymbol{Q}(t)

under this S-only algorithm is the same as the unconditional expectation (because S(t) is i.i.d. over slots, and the S-only algorithm is independent of current queue backlogs) yields:

\Delta(t)\leqB

Thus, the drift of a quadratic Lyapunov function is less than or equal to a constant B for all slots t. This fact, together with the assumption that queue arrivals have bounded second moments, imply the following for all network queues:[16]

\limt → infty

(c)
Q(t)
n
t

=0withprobability1

For a stronger understanding of average queue size, one can assume the arrival rates

(c)
(λ
n

)

are interior to

Λ

, so there is an

\epsilon>0

such that Eq. (9) holds for some alternativeS-only algorithm. Plugging Eq. (9) into the right-hand-side of Eq. (10) yields:

\Delta(t)\leqB-

(c)
\epsilon\sum
n

(t)

from which one immediately obtains (see[2] [11]):

\limsupt → infty

1
t
t-1
\sum
\tau=0
N
\sum
n=1
N
\sum
c=1

E\left[

(c)
Q
n

(\tau)\right]\leq

B
\epsilon

This average queue size bound increases as the distance

\epsilon

to the boundary of thecapacity region

Λ

goes to zero. This is the same qualitative performance as a single M/M/1 queue with arrival rate

λ

and service rate

\mu

, whereaverage queue size is proportional to

1/\epsilon

, where

\epsilon=\mu

.

Extensions of the above formulation

Non-i.i.d. operation and universal scheduling

The above analysis assumes i.i.d. properties for simplicity. However, the same backpressure algorithm can be shown to operate robustly in non-i.i.d. situations. When arrival processes and topology states are ergodic but not necessarily i.i.d., backpressure still stabilizes the system whenever

(c)
(λ
n

)\inΛ

.[7] More generally, using a universal scheduling approach, it has been shown to offer stability and optimality properties for arbitrary (possibly non-ergodic) sample paths.[17]

Backpressure with utility optimization and penalty minimization

Backpressure has been shown to work in conjunction with flow control via a drift-plus-penalty technique.[8] [9] [2] This technique greedily maximizes a sum of drift and a weighted penalty expression. The penalty is weighted by a parameter V that determines a performance tradeoff.This technique ensures throughput utility is within O(1/V) of optimality while average delay is O(V). Thus, utility can be pushed arbitrarily close to optimality, with a corresponding tradeoff in average delay. Similar properties can be shown for average power minimization[18] and for optimization of more general network attributes.[11]

Alternative algorithms for stabilizing queues while maximizing a network utility have been developedusing fluid model analysis,[10] joint fluid analysis and Lagrange multiplier analysis,[19] convex optimization,[20] and stochastic gradients.[21] These approaches do not provide the O(1/V), O(V) utility-delay results.

See also

Primary sources

Notes and References

  1. L. Tassiulas and A. Ephremides,"Stability Properties of Constrained Queueing Systems andScheduling Policies for Maximum Throughput in MultihopRadio Networks, IEEE Transactions on Automatic Control, vol. 37, no. 12, pp. 1936-1948, Dec. 1992.
  2. L. Georgiadis, M. J. Neely, and L. Tassiulas, "Resource Allocation and Cross-Layer Control in Wireless Networks,"Foundations and Trends in Networking, vol. 1, no. 1, pp. 1-149, 2006.
  3. L. Jiang and J. Walrand. Scheduling and Congestion Control for Wireless and Processing Networks,Morgan & Claypool, 2010.
  4. A. Sridharan, S. Moeller, and B. Krishnamachari,"Making Distributed Rate Control using Lyapunov Drifts a Reality in Wireless Sensor Networks,"6th Intl. Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks (WiOpt),April 2008.
  5. A. Warrier, S. Janakiraman, S. Ha, and I. Rhee, "DiffQ: Practical Differential Backlog Congestion Control for WirelessNetworks," Proc. IEEE INFOCOM, Rio de Janeiro, Brazil, 2009.
  6. B. Awerbuch and T. Leighton, "A Simple Local-Control Approximation Algorithm for Multicommodity Flow," Proc. 34th IEEE Conf.on Foundations of Computer Science, Oct. 1993.
  7. M. J. Neely, E. Modiano, and C. E. Rohrs, "Dynamic Power Allocation and Routing for Time Varying Wireless Networks," IEEE Journal on Selected Areas in Communications, vol. 23, no. 1, pp. 89-103, January 2005.
  8. M. J. Neely. Dynamic Power Allocation and Routing for Satellite and Wireless Networks with Time Varying Channels.Ph.D. Dissertation, Massachusetts Institute of Technology, LIDS. November 2003.
  9. M. J. Neely, E. Modiano, and C. Li, "Fairness and Optimal Stochastic Control for Heterogeneous Networks," Proc. IEEE INFOCOM, March 2005.
  10. A. Stolyar,"Maximizing Queueing Network Utility subject to Stability: Greedy Primal-Dual Algorithm,"Queueing Systems,vol. 50, no. 4, pp. 401-457, 2005.
  11. M. J. Neely.Stochastic Network Optimization with Application to Communication and Queueing Systems,Morgan & Claypool, 2010.
  12. M. J. Neely and R. Urgaonkar, "Optimal Backpressure Routing in Wireless Networks with Multi-Receiver Diversity," Ad Hoc Networks (Elsevier), vol. 7, no. 5, pp. 862-881, July 2009.
  13. S. Moeller, A. Sridharan, B. Krishnamachari, and O. Gnawali,"Routing Without Routes: The Backpressure Collection Protocol,"Proc. 9th ACM/IEEE Intl. Conf. on Information Processing in Sensor Networks (IPSN),April 2010.
  14. L. Huang, S. Moeller, M. J. Neely, and B. Krishnamachari, "LIFO-Backpressure Achieves Near Optimal Utility-Delay Tradeoff,"Proc. WiOpt, May 2011.
  15. E. Modiano, D. Shah, and G. Zussman, "Maximizing throughput in wireless networks via gossiping," Proc. ACM SIGMETRICS, 2006.
  16. M. J. Neely, "Queue Stability and Probability 1 Convergence via Lyapunov Optimization," Journal of Applied Mathematics, vol. 2012, .
  17. M. J. Neely, "Universal Scheduling for Networks with Arbitrary Traffic, Channels,and Mobility," Proc. IEEE Conf. on Decision and Control (CDC), Atlanta, GA, Dec. 2010.
  18. M. J. Neely, "Energy Optimal Control for Time Varying Wireless Networks,"IEEE Transactions on Information Theory, vol. 52, no. 7, pp. 2915-2934,July 2006
  19. A. Eryilmaz and R. Srikant, "Fair Resource Allocation in Wireless Networks using Queue-Length-Based Schedulingand Congestion Control," Proc. IEEE INFOCOM, March 2005.
  20. X. Lin and N. B. Shroff, "Joint Rate Control and Scheduling in Multihop Wireless Networks,"Proc. of 43rd IEEE Conf. on Decision and Control, Paradise Island, Bahamas, Dec. 2004.
  21. J. W. Lee, R. R. Mazumdar, and N. B. Shroff, "Opportunistic Power Scheduling for Dynamic Multiserver Wireless Systems,"IEEE Transactions on Wireless Communications, vol. 5, no.6, pp. 1506–1515, June 2006.