Documente Academic
Documente Profesional
Documente Cultură
Go TO Answer Directly
info@statisticsassignmenthelper.com ESD.86
Massachusetts Institute
Problem Setof#4Technology
2. Squaring. We’ll be doing lots of squaring of random variables in the 2nd half of this
subject. So, let’s get started now.
(a) Suppose we know all about random variable X, having pdf
f X (x) for all x,"# < x < +# . Define new r.v. Y=X2. Show that the following is
1
true: fY (y) = [ f X ( y ) + f X (" y )] . Explain in words what this means.
2 y
! (b) Apply your result to a situation in which X is uniformly distributed over the
interval [-1, +2].
!
3. M/M/k Queue. Let’s examine the queue with homogeneous Poisson arrivals (rate λ),
negative exponential service times (mean 1/µ), k identical servers, infinite queue capacity
and no balking or reneging. We assume the steady state exists, implying that λ< kµ. Let
the state of the system be the number of customers in the system, both in service and in
queue.
(a) Find the steady state probability Pn=P{n customers in the system}, n=0,1,2,…
(b) Find W. (Hint: think of Little’s Law and that finding any one of 4 quantities gets
you the other 3 at ‘no extra charge.’ So, to find W all you need to do is compute
the easiest one of the 4 quantities related to Little’s Law.)
(c) The systems designers of call centers and other systems involving queueing like
to enjoy economies of scale. Yet they have service performance standards to
meet. To many in the services industries, the quantity (1-ρ) is ‘paid lost time’,
that is, time that a server is not working serving customers. Suppose a service
system that can be modeled as an M/M/k Queue has a performance standard that
states that the mean wait in queue shall be no greater than one minute. Assume
the mean service time is 1/µ = one minute. Using your results above and an
Excel or similar spreadsheet, examine how (1-ρ) can be made smaller as the call
volume λ increases, the number of servers k increases, while at all times meeting
the service standard of one minute or less mean queue wait time. Show that (1-ρ)
https://www.statisticsassignmenthelper.com/ 1
can become much smaller than the value associated with one server as call
volume increases, the number of servers increases, but not in direct proportion to
call volume.
(d) Suppose you designed your system as in Part (c), and you have 20 servers in your
system. How sensitive are you to errors in forecasted call volumes? What
happens if tomorrow’s call volume is 10% over forecasts? This is a matter of
‘ilities,” reliability, robustness, resilience. Did you follow in the media what
happened to stock traders and brokers and even stock indices as trading volume
zoomed far above routine averages on Tuesday February 27?
4. M/M/k Queue with Discouraged Voters. We now want to look at a variation of the
queue of problem 3. Suppose it is Election Day and we focus on a given voting precinct.
Systems managers have a choice before Election Day: how many voting machines
(servers) to place in each precinct. Too many in one precinct may mean no queues at that
precinct but that another precinct is short-changed. Too few implies long lines of voters,
some of whom may not join the line because it is too long.
(a) Suppose arrivals that see a line of queued voters can get discouraged and
not join the line. Suppose field research has shown that the probability
that a potential voter will join the system and vote is equal to
"0.1n
pV (n q ) = e q , where n q # number of voters in queue upon arrival, n q = 0,1,2,...
.
For simplicity we assume no reneging, that is, once someone joins the
voters’ line, they stay and eventually cast their vote. Model this system as
! a Markov Birth and Death Queueing system. Find the steady state
probabilities, i.e.,
Pn " {n voters in the system, both in queue and in voting booths}, n = 0,1,2,...
2
Problem Set #4
1 "
Little’s Law: L = λ.W and Lq = λ.Wq, with λ = λL + λR, W = + W q and L = Lq + .
u u
T T
Rule A: W A = " LA = ( # L + # R )
2 2
N 0 "1 N 0 "1 ! !
Rule B: W B = $ LB =
2( #L + #R ) 2
!Rule C: W = T0 #1+ 1 & ( "L + "R ).T0 # 1 &
C % ( ) LC = %1+ (
2 $ 1+ ( "L + "R ).T0 ' 2 $ 1+ ( "L + "R ).T0 '
!
2. Squaring – 4 pts
a) 2 pts
!
FY (y) = P[Y " y] = P[X 2 " y] = P # y " X " y [ ]
Let FY(y) be the CDF of fY(y): y
FY (y) = $ f X (x).dx = FX ( y ) #F (# y )
X
# y
d
Then, fY (y) =
dy
[ ( y ) " F (" y )] = 2.1 y [ f ( y ) + f (" y )]
FX X X X
!
b) 2 pts
1
! f X (X) = ,"X # [$1;2]
3
% 1
'' 3. y ,#y $ [0;1]
1
!
Therefore, fY (y) =
2. y
[ f ( y ) + f (" y )]
X X =&
' 1 ,#y $ [1;4]
(' 6. y
!
3. M/M/k Queue – 6.5 pts
a) 2 pts
- # " &n
/% (
/ $ µ ' P ,)n * [0;k +1]
/ 0
P(n) = . n! n
/ #"&
/ %$ µ ('
/ n+k P0 ,)n * [k;,[
0 k .k!
- $ # 'n $ # 'n 0
,
/ k*1 & ) , & )
2
/ %µ( %µ( 2 1
Since + Pn = 1 " P0 + + + n*k = 1, we have: P0 =
! / n! k .k!2 #"& n
# " &n
n= 0 n= 0 n= k
/ 2 k)1 % ( + % (
$µ' µ
. 1
* n! + * k$n)k'.k!
n= 0 n= k
This becomes:
1 1 1
! P0 = n n
= n
= n
#"& # "& # "& #"&
% ( !
k)1 + % ( k)1 % ( k + m k)1 % ( k
$µ' µ $µ' # "& 1 # " & $µ' # "& 1 1
* n! + * k$n)k'.k! * n! + %$ µ (' k! , *%$ µ.k (' * n! + %$ µ (' k! , "
n= 0 n= k n= 0 m= 0 n= 0 1)
µ.k
1
P0 = n
#"&
k)1 % (
$µ' "k
* n! µ k)1.(µ.k ) ")(k )1)!
+
n= 0
We can then deduce the exact value for each Pn.
b) 1.5 pt
! %
#
& % )k 1 # & % )m & % )k 1 k.µ & % ) k +1 P0
Lq = $ (n " k).Pn =P0 ( + , $ m( + = P0 ( + , 2 =( + 2
n= k ' µ * k! m= 0 ' k.µ * ' µ * k! & % ) 'µ* & % )
(1" + k.k!(1" +
' k.µ * ' k.µ *
1 L 1
Since W = W q + = q + ,
µ " µ
k
! " .P0 1
W = 2 +
$ " ' µ
µ k +1 .k.k!&1# )
! % k.µ (
!
c) 1.5 pts
A simulation shows that, when fixing Wq = 1 and 1/µ = 1 min, the progression of the maximum
number of calls is faster than the progression of the number of servers required. In addition, the
d) 1.5 pts
In such a case, the service quality is extremely sensitive to volume fluctuation. A mere 7%
increase in calls and the call center can no longer service customers fast enough: the expected
On February 27th, trading volumes exceeded the trading capacity of the stock market, creating an
enormous queue of transactions. The value of stocks lagged far behind, creating confusion
among traders.
0 n+k P , e 20
,)n * [k;-[
1 k .k! 0
. $ # 'n $ #'
n 1
,
0 k*1 & ) , & ) (n*k )(n*k +1)
3
! 0 %µ( %µ( * 3 =1
+ Pn = 1 " P00+ n! + + k n*k .k! - e 20
3
n= 0
0n= 0 n= k
3
/ 2
1 1
This becomes: P0 = n
=
#"& # " &n n
#"&
! % (
k)1 , % ( (n)k )(n)k +1) k)1 % ( # &k m m(m +1)
µ $µ' $µ' " 1 ,# " &
* $ n!'
) )
+ * n)k + e 20
* n! + %$ µ (' k! + *%$ µ.k (' + e 20
n= 0 n= k k .k! n= 0 m= 0
b) 1.5 pts
!# k%1 #
%
n%k
"Av = $ "n .Pn = $ " .Pn + $ " .e 10 .Pn
n= 0 n= 0 n= k
!
$ $
L
% (n # k).P n % m.P m +k
In steady state, we get: W q = q = k#1
n= k
$ n#k = m= 0
"Av # & k#1 $
#
m )
% ".P + % ".e
n
10
.Pn "(% Pn + % e 10 .Pm +k +
n= 0 n= k 'n= 0 m= 0 *
c) 1.5 pt
No, it’s not. People would balked out do are not taken into account here. The quality of service
!
will be overestimated.
d) 1.5 pt
#Av & k"1 $
"
m )
Percentage of people who balked: PBal = 1" = 1" (% Pn + % e 10 .Pm +k +
# 'n= 0 m= 0 *