20.11.2012 Views

Contents Telektronikk - Telenor

Contents Telektronikk - Telenor

Contents Telektronikk - Telenor

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

220<br />

Notes on a theorem of L. Takács<br />

on single server queues with feedback<br />

BY ELIOT J JENSEN<br />

Abstract<br />

L. Takács published in 1963 a paper<br />

introducing a queuing model with<br />

feedbacks. Such models may be used to<br />

calculate the performance of certain<br />

telecommunication processor systems.<br />

This note proposes alternative formulations<br />

of some of the results in the<br />

paper. They may be convenient when<br />

calculating response times for strings<br />

of processing tasks representing e.g.<br />

call handling functions. The impact of<br />

task scheduling on the range of<br />

response times has been indicated.<br />

Background<br />

L. Takács [1] examines the system<br />

sojourn times of tasks offered to a single<br />

server, with an infinite queue in front of<br />

it. Tasks arrive to the system either from<br />

outside, according to a Poisson process<br />

with intensity λ, or as feedbacks. A feedback<br />

may be generated at the termination<br />

of the service of a task, and immediately<br />

put at the tail of the queue. The probability<br />

for this is p and for no feedback<br />

q =1–p. Tasks are assumed to have<br />

independent but identically distributed<br />

service times. This distribution is<br />

assumed to have the Laplace-Stieltje<br />

transform Ψ(s). Let its mean value be α.<br />

Theorem 3 of the Takács paper considers<br />

the stationary distribution of the total<br />

sojourn time θn (n = 1, 2, ...) for a random<br />

task and its feedbacks, and reads,<br />

cit:<br />

If λα < q, then θn has a unique stationary<br />

distribution P{θn ≤x},<br />

which is<br />

given by the following Laplace-Stieltje<br />

transform<br />

∞�<br />

Φ(s) =q p (1)<br />

k−1 Uk(s, 1) (R(s) ≥ 0)<br />

where<br />

k=1<br />

U1 (s,z) = P0Ψ(s + λ(1 – z))<br />

+ U(s + λ(1 – z),<br />

(q + pz)Ψ(s + λ(1 – z)))(2)<br />

for R(s) ≥0<br />

and |z| ≤1,<br />

P0 = 1 – λα/q,<br />

U(s,z) is defined by (20) of [1], and<br />

Uk+1 (s,z) = Ψ(s + λ(1 – z))<br />

Uk (s,(q + pz)Ψ<br />

(s + λ(1 – z))) (3)<br />

for k = 1, 2, ...<br />

The formula U(s,z) (eq. 20 of [1]) is defined<br />

as the combined Laplace-Stieltje<br />

transform and z-transform of a distribution<br />

P j (t), giving the probability of, in<br />

the stationary state (assuming λα < q), to<br />

encounter exactly j (> 0) tasks in the system,<br />

where the task in the server has<br />

remaining service time shorter or equal<br />

to t. It reads:<br />

U(s, z) =<br />

�<br />

1 − λα<br />

�<br />

q<br />

(4)<br />

λz(1 − z)(Ψ(s) − Ψ(λ(1 − z)))<br />

(z − (q + pz)Ψ(λ(1 − z)))(s − λ(1 − z))<br />

We also note that Uk (s,z) is the Laplace-<br />

Stieltje transform and z-transform of the<br />

combined distribution of the total sojourn<br />

time of a task and its feedbacks, and the<br />

number of tasks left behind when the last<br />

of its feedbacks leaves the server, provided<br />

the number of feedbacks is exactly<br />

k – 1 for that particular task.<br />

From theorem 3 is obtained<br />

∞�<br />

Φ(s, z) =q p k−1 Uk(s, z) =<br />

k=1<br />

qU1(s, z)+pΨ(s + λ(1 − z))<br />

Φ(s, (q + pz)Ψ(s + λ(1 − z)))<br />

Then the moments<br />

E {θ r n} =(−1) r<br />

� �<br />

r ∂<br />

Φ(s, z)<br />

∂sr for the stationary sojourn time process<br />

may be found by solving a set of r +1<br />

linear equations for the determination of<br />

Φij =<br />

� �<br />

i+j ∂ Φ(s, z)<br />

∂s i ∂z j<br />

Takács has given explicit formulae for<br />

E{θ n r } for r = 1 and 2.<br />

(5)<br />

(6)<br />

An alternative formulation<br />

In the recurrence formula for Uk (s,z) in<br />

theorem 3, make the substitution<br />

ξ1 (s,z) = (q + pz)Ψ(s + λ(1 – z)) (7)<br />

and introduce the recurrence<br />

( s=0<br />

z=1)<br />

( s=0<br />

z−1)<br />

ξk (s,z) = (q + pξk–1 (s,z))<br />

Ψ(s + λ(1 – ξk–1 (s,z))) (8)<br />

starting with ξ 0 (s,z) = z. Then, by rearrangement<br />

we obtain<br />

Uk+1(s, z) =<br />

� k−1<br />

U1(s, ξk(s, z))<br />

(k = 0, 1, 2, ...)<br />

(9)<br />

It is easily seen that ξℓ(s, z) is the<br />

Laplace-Stieltje transform and z-transform<br />

of the combined distribution of system<br />

sojourn time and the number of tasks<br />

left in the system for a task which needs<br />

to enter the server ℓ times, but always<br />

with zero service time, the task joining<br />

the system finding only one (other) task<br />

there, which is just about to enter service.<br />

It follows that<br />

for |z| ≤1<br />

and λα < q.<br />

�<br />

�<br />

Ψ(s + λ(1 − ξℓ(s, z)))<br />

ℓ=0<br />

lim<br />

ℓ→∞ ξℓ(s, z) = ¯ B(s)<br />

¯B(s) is the Laplace-Stieltje transform of<br />

the busy period distribution.<br />

For convenience the formula for Uk (s,z)<br />

may be slightly rewritten, using the ztransform<br />

and the Laplace-Stieltje transform<br />

of the number of tasks in the system<br />

and the remaining service time of the<br />

task in the server for the stationary process<br />

at a random time. If the system is<br />

empty, the remaining service time is zero.<br />

Based on theorem 1 of [1] we may put<br />

�<br />

P (s, z) = 1 − λα<br />

�<br />

q<br />

�<br />

λ(1 − z)<br />

1+z<br />

s − λ(1 − z)<br />

�<br />

Ψ(s) − Ψ(λ(1 − z))<br />

·<br />

z − (q + pz)Ψ(λ(1 − z))<br />

(10)<br />

A task (which we may call a tagged task)<br />

arriving at a random instant to the system<br />

will see this distribution. When the<br />

remaining service time finishes, a feedback<br />

may, or may not be generated. At<br />

the moment the server is about to start<br />

processing the next task, the tagged task<br />

will see a number i of tasks ahead of it in<br />

the queue and a number j of tasks behind,<br />

and at that time it will have waited a certain<br />

time t (identical to the remaining service<br />

time of the task initially in the<br />

server). The Laplace-Stieltje and z-transform<br />

of the corresponding distribution is<br />

�<br />

Q(s, y, z) = 1 − λα<br />

�<br />

(1 + (q + pz)<br />

q<br />

λ(1 − y)<br />

s + λ(1 − z) − λ(1 − y) ·<br />

�<br />

Ψ(s + λ(1 − z)) − Ψ(λ − y))<br />

y − (q + py)Ψ(λ(1 − y))<br />

(11)

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!