Advanced Queueing Theory • Networks of queues (reversibility, output theorem, tandem networks, partial balance, product-form distribution, blocking, insensitivity, BCMP networks, mean-value analysis, Norton's theorem, sojourn times) • Analytical-numerical techniques (matrix-analytical methods, compensation method, error bound method, approximate decomposition method) • Polling systems (cycle times, queue lengths, waiting times, conservation laws, service policies, visit orders) Richard J. Boucherie department of Applied Mathematics University of Twente http://wwwhome.math.utwente.nl/~boucherierj/onderwijs/Advanced Queueing Theory/AQT.html 1 Last time on AQT … • • • • • • Output reversible queue Tandem network of queues Jackson network Kelly Whittle network Partial balance Blocking 2 Advanced Queueing Theory 3 Today (lecture 3): queue length based on Kelly’s lemma Nelson, sec 10.4—10.7, Kelly, chapter 3 • Kelly Whittle network • Alternative proof • Quasi reversibility • Network of quasi reversible queues • Queue disciplines, symmetric queues, BCMP networks • PASTA • Mean Value Analysis Kelly Whittle network Theorem: The equilibrium distribution for the Kelly Whittle network is where and π satisfies partial balance 4 5 Insert equilibrium distribution and rates in partial balance This is the beauty of partial balance! rate in = rate out Advanced Queueing Theory 6 Today (lecture 3): queue length based on Kelly’s lemma Nelson, sec 10.4—10.7, Kelly, chapter 3 • Kelly Whittle network • Alternative proof • Quasi reversibility • Network of quasi reversible queues • Queue disciplines, symmetric queues, BCMP networks • PASTA • Mean Value Analysis Time reversed process 7 • Theorem: If X(t) is a stationary Markov process with transition rates q(j,k), and equilibrium distribution π(j), j ∈ S, then the reversed process X(τ - t) is a stationary Markov process with transition rates and the same equilibrium distribution. € € • Theorem: Kelly’s lemma Let X(t) be a stationary Markov process with transition rates q(j,k). If we can find a collection of numbers q’(j,k) such that q’(j)=q(j), j∈S, and a collection of positive numbers π (j), j ∈ S, summing to unity, such that € then q’(j,k) are the transition rates of the time-reversed process, and π (j), j∈S, is the equilibrium distribution € €of both processes. 8 Alternative proof: use Kelly’s lemma Forward rates Guess backward rates Check conditions 9 Alternative proof: use Kelly’s lemma Guess for equilibrium distribution insert Advanced Queueing Theory 10 Today (lecture 3): queue length based on Kelly’s lemma Nelson, sec 10.4—10.7, Kelly, chapter 3 • Kelly Whittle network • Alternative proof • Quasi reversibility • Network of quasi reversible queues • Queue disciplines, symmetric queues, BCMP networks • PASTA • Mean Value Analysis Quasi-reversibility • Multi class queueing network, class c∈ C • A queue is quasi-reversible if its state x(t) is a stationary Markov process with the property that the state of the queue at time t0, x(t0), is independent of € (i) arrival times of class c customers subsequent to time t0 (ii) departure times of class c customers prior to time t0. • Theorem If a queue is QR then (i) arrival times of class c customers form independent Poisson processes (ii) departure times of class c customers form independent Poisson processes. Quasi-reversibility • S(c,x) set of states queue contains one more class c customer than in state x • Arrival rate class c customer independent of state x • Thus arrival rate independent of prior events, and has constant rate Poisson proces Quasi-reversibility • Multi class queueing network, class c∈ C • S(c,x) set of states in which queue contains one more class c than in state x • Arrival rate class c customer independent of state x € • Departure rate class c customer independent of state x • Characterise QR, combine • Partial balance Advanced Queueing Theory 14 Today (lecture 3): queue length based on Kelly’s lemma Nelson, sec 10.4—10.7, Kelly, chapter 3 • Kelly Whittle network • Alternative proof • Quasi reversibility • Network of quasi reversible queues • Queue disciplines, symmetric queues, BCMP networks • PASTA • Mean Value Analysis 15 Network of quasi reversible queues • • • • • • • Multiclass queueing network, type i=1,..,I J queues Customer type identifies route Poisson arrival rate per type ν(i), i=1,…,I Route r(i,1),r(i,2),…,r(i,S(i)) Type i at stage s in queue r(i,s) S(c,x) set of states in which queue contains one more class c than in state x • State X(t)=(x1(t),…,xJ(t)) • Fixed number of visits; cannot use Markov routing • 1, 2, or 3 visits to queue: use 3 types Network of quasi reversible queues • Construct network by multiplying rates for individual queues • Transition rates • Arrival of type i causes queue k=r(i,1) to change at • Departure type i from queue j = r(i,S(i)) • Routing • Internal 16 Network of Quasi-reversible queues • Rates • Theorem : For an open network of QR queues (i) the states of individual queues are independent at fixed time (ii) an arriving customer sees the equilibrium distribution (ii’) the equibrium distribution for a queue is as it would be in isolation with arrivals forming a Poisson process. (iii) time-reversal: another open network of QR queues (iv) system is QR, so departures form Poisson process Network of Quasi-reversible queues • Proof of part (i): Kelly’s lemma • Rates • Transition rates reversed process (guess) Network of Quasi-reversible queues • Proof of part (i): Kelly’s lemma • For • We have • Satisfied due to Advanced Queueing Theory 20 Today (lecture 3): queue length based on Kelly’s lemma Nelson, sec 10.4—10.7, Kelly, chapter 3 • Kelly Whittle network • Alternative proof • Quasi reversibility • Network of quasi reversible queues • Queue disciplines, symmetric queues, BCMP networks • PASTA • Mean Value Analysis Queue disciplines • Operation of the queue j: (i) Each job requires exponential(1) amount of service. (ii) Total service effort supplied at rate ϕj(nj) (iii) Proportion γ j(k,nj) of this effort directed to job in position k, k=1,…, nj ; when this job leaves, his service is completed, jobs in positions k+1,…, nj move to positions k,…, nj -1. (iv) When a job arrives at queue j he moves into position k with probability δj(k,nj + 1), k=1,…, nj +1; jobs previously in positions k,…, nj move to positions k+1,…, nj +1. Queue disciplines • Operation of the queue j: (i) Each job requires exponential(1) amount of service. (ii) Total service effort supplied at rate ϕj(nj) (iii) Proportion γ j(k,nj) of this effort directed to job in position k, (iv) job arriving at queue j moves into position k with prob. δj(k,nj + 1) • In the form of network of quasi reversible queues • Examples: BCMP network FCFS LCFS PS infinite server queue, Symmetric queues • Operation of the queue j: (i) Each job requires exponential(1) amount of service. (ii) Total service effort supplied at rate ϕj(nj) (iii) Proportion γ j(k,nj) of this effort directed to job in position k, (iv) job arriving at queue j moves into position k with prob. δj(k,nj + 1) • • • • Examples: IS, LCFS, PS Symmetric queue QR for general service requirement Instantaneous attention Symmetric queue is insensitive Quasi-reversibility and partial balance • QR: fairly general queues, service disciplines, Markov routing, product form equilibrium distribution factorizes over queues. • PB: fairly general relation between service rate at queues, state-dependent routing (blocking), product form equilibrium distribution factorizes over service and routing parts. • Identical for single type queueing network with Markov routing • • • • QR partial balance NOT partial balance QR (exercise) NOT QR Reversibility (see Nelson, ex 10.14) NOT Reversibility QR (see Nelson, ex 10.12) Advanced Queueing Theory 25 Today (lecture 3): queue length based on Kelly’s lemma Nelson, sec 10.4—10.7, Kelly, chapter 3 • Kelly Whittle network • Alternative proof • Quasi reversibility • Network of quasi reversible queues • Queue disciplines, symmetric queues, BCMP networks • PASTA • Mean Value Analysis Arrival theorem • PASTA: The distribution of the number of customers in the system seen by a a customer arriving to a system according to a Poisson process (i.e., at an arrival epoch) equals the distribution of the number of customers at an arbitrary epoch. • Arrival theorem (open Jackson network): In an open network in equilibrium, a customer arriving to queue j observes the equilibrium distribution. • Arrival theorem (closed Jackson network): In a closed networkin equilibrium, a customer arriving to queue j observes the equilibrium distribution of the network containing one customer less. 27 PASTA: Poisson Arrivals See Time Averages • fraction of time system in state n • probability outside observer sees n customers at time t • probability that arriving customer sees n customers at time t (just before arrival at time t there are n customers in the system) • in general PASTA: Poisson Arrivals See Time Averages 28 • Let C(t,t+h) event customer arrives in (t,t+h) • For Poisson arrivals q(n,n+1)=λ so that • Alternative explanation; PASTA holds in general! PASTA: Poisson Arrivals See Time Averages 29 • Transient • In equilibrium 30 MUSTA: Moving Units See Time Averages • Palm probabilities: Each type of transition nnʼ for Markov chain associated with subset H of SxS \diag(SxS) • Example:transition in which customer queue i queue j H ij = {(m + ei ,m + e j ),m + ei ,m + e j ∈ S} m • Transition in which customer leaves queue i H out = H ij i j • Transition in which customer enters queue j H inj = H ij i 31 MUSTA: Moving Units See Time Averages • NH process counting the H-transitions • Palm probability PH (C) of event C given that H occurs: PH ∑ (C) = ∑ (n,n' )∈C (n,n' )∈ H • π (n)q(n,n') π (n)q(n,n') , C⊆H Probability customer queue i queue j sees state m Pij (m) = PH ij ((m + ei ,m + e j )) = π (m + ei )q(m + ei ,m + e j ) ∑ (n,n' )∈ H ij • π (n)q(n,n') , Probability customer arriving to queue j sees state m P j (m) = PH in j π (m + e )q(m + e ,m + e ) ∑ ( (m + e ,m + e )) = , π (n)q(n,n') ∑ ∑ i i i i j i (n,n' )∈ H ij i j Kelly Whittle network Theorem: The equilibrium distribution for the Kelly Whittle network is where and π satisfies partial balance 32 MUSTA :Kelly Whittle network Theorem: The distribution seen by a customer moving from queue i tot queue j is Entering queue j is where 33 MUSTA 34 Advanced Queueing Theory 35 Today (lecture 3): queue length based on Kelly’s lemma Nelson, sec 10.4—10.7, Kelly, chapter 3 • Kelly Whittle network • Alternative proof • Quasi reversibility • Network of quasi reversible queues • Queue disciplines, symmetric queues, BCMP networks • PASTA • Mean Value Analysis Gesloten netwerken: MVA Average queue length, average sojourn times? λm(i) arrival intensity queue i, Fm(i) expeceted sojourn time i, Lm(i) expected queue length queue i, when m cust in system Arrival theorem, FCFS Little’s formula Gesloten netwerken: MVA λm(i) arrival intensity queue i, Fm(i) expeceted sojourn time i, Lm(i) expected queue length queue i, when m cust in system N N π i = ∑ π j ⋅ rji λm (i) = ∑ λm ( j) ⋅ rji j=1 N N j=1 € i=1 € € • thus N λm = m ⋅ ∑ π i ⋅ Fm (i) i=1 ∑π i=1 λm (i) = λm ⋅ π i , λm = ∑ λm (i) • Little , −1 i =1 • Mean Value Analysis evaluates λm(i), Fm(i) en Lm(i) for all m,i recursively – Find solution π of traffic equations – for m=1 : F1(i)=1/µi for all i recursion – let Fm(i) known for all i – Determine number of cust served per time unit at queue i : λm (i) = λm ⋅ π i = m ⋅ {∑ N j=1 −1 } π j ⋅ Fm ( j) ⋅ πi – Determine average number of customers at queue i using Little Lm (i) = λm (i)⋅ Fm (i) – Determine average sojourn time at queue i for system €containing m+1 customers using arrival theorem 1+ Lm (i) Fm +1 (i) = µi € Exercises Exercises from Kelly, Reversibility and stochastic networks: 1.5.2, 1.6.4, 2.2.4, 2.3.5, 2.4.1, 3.1.3, 3.1.4, 3.2.3 Exercise on blocking: Consider a Jackson network of J queues with transition rates 1. Give the equilibrium distribution, and show that this distribution satisfies partial balance 2. Show that a customer arriving to queue i observes the equilibrium distribution 3. Now assume that the total number of customer in the network is restricted not to exceed N. Give the equilibrium distribution, and the distribution of the state observed by a customer arriving to queue 1 4. Now assume that each queue has a capacity restriction, say the number in queue i is restricted not to exceed Ni. Consider the stop protocol, where the arrival process, and each non-blocked queue is stopped until a service completion has occurred at the blocked queue. Give the state space, and show that the equilibrium distribution has a product form and satisfies partial balance. Give the distribution of the state observed by a customer arriving to (and entering) queue 1. 5. Again assume the capacity restrictions of 4). Assume that a customer arriving to a queue that has reached its capacity restriction, say queue j, jumps over the queue, and selects a new destination according to the routing probilities from queue j. Give the equilibrium distribution. 6. Can the product form result of 4) be obtained via Quasi reversibility? If so, provide the proof of the product form result via quasi reversibility. Exercise on insensitivity: Consider a BCMP network consisting of infinite server queues, and single server queues, where the single server queues may be FCFS, LCFS or PS. Assume Markov routing. 1. Formule the BCMP network as a network of quasi reversible queues. 2. Give the equilibrium distribution. 3. Show that this distribution satisfies partial balance. 4. Restrict the network to contain IS, LCFS, and PS queues only (symmetric queues). Prove that the equilibrium is insensitive to the distibution of the service time at each of these queues using first the phase type distribution approach, and then the formulation of Whittle. 5. Indicate why the distribution cannot be insensitive to the service time distribution in an FCFS queue.