ANALYSIS AND OPTIMIZATION OF MARKOV HM-NETWORKS WITH STOCHASTIC INCOMES FROM TRANSITIONS BETWEEN THEIR STATES
Michal Matalycki1, Katarzyna Koluzaewa2
1Institute of Mathematics, Czestochowa University of Technology, Poland
2Faculty of Mathematics and Computer Science, Grodno State University, Bielarus koluzaeva@gmail.com
Abstract. Expressions for expected incomes and variations of incomes in systems of Mar- kov HM-networks, when service rates of messages are depending or not depending of network’s states are obtained. The case when incomes of transitions between network’s states are random variables with the set moments of first two orders is considered. Solution of some optimization problems for HM-networks are resulted.
Introduction
It is known that functioning of any Markov queueing network (QN) is described with a help of Markovian chain with continuous time. In Howard’s work [1] the conception of Markovian chains with incomes that were constants was introduced and there were proposed to use method of Laplace transformation and method of z-transformation for analysis of such chains with small number of states. This con- cept has laid down in a basis of definition of Markov QN with incomes that were examined in works [2-4] at first. Later open and closed networks with central queue- ing system (QS) were investigated in cases when a) incomes from transitions be- tween network’s states depend on states and time [5-7] or b) incomes are random variables (RV) with the set moments of the first and the second orders [8, 9]. Re- cently QN with incomes refer to as HM (Howard-Matalytski)-networks. In the present article expressions in case b) were obtained for expected incomes and in- come’s variations in systems of open exponential HM-networks and Jackson HM- networks of arbitrary architecture.
Let us examine open exponential QN with one type messages that consist of n QS S1,S2,...,Sn. Vector k(t) (k,t)
k1,k2,...,kn,t is state of network where ki is number of messages in system S at the momenti t,t t[ 0,f), i 1 n, . For un- ification of designation let us introduce system S (outside medium) from which the 0 Poisson flow of messages with arrival rate O comes into network. At first we willexamine the case when service rate of messages Pi in system S doesn’t depend on i
number of messages in it, i 1,n. Let p0j is probability of message entry from system S to system0 Sj, 1;
1
¦
n 0 jp j p is probability of message transition after ij
its service in system S to system i Sj, 1,
¦
n0 jpij i 1 n, .Message during its transi- tion from system S to systemi Sj brings to system Sj some random income and income of system S descend on this value correspondently,i i,j 0,n. It is needed to find expected (average) incomes of network’s systems during time t considering that network’s state in the starting time t0 is known. At first we will suppose that network is functioning in the term of high capacity, i.e. ki(t)!0, t[t0,f),
. , 1 n i
1. Expected incomes of network systems
Let us examine dymamics of income’s change of some network’s system Si. Let us denote its income at the moment t as Vi(t). The income of this system at some moment t is equal to 0 vi0. We will be interested in system’s incomeVi(t0 t) at the moment t0 To a finding of the income we shall apply a following technique. Let t. us divide interval
>
t0,t0t@
into m equal portions with length ǻ ,m
t t and consid- er that m is big. For finding of system’s income Let us write probability of all events that can occur at the l -th time interval, l 1 m, . The next situations are poss- ible: a) message from outside will arrive to system Si with probability
ǻ ) ǻ (
0 t o t
p i
O and will bring income of size r0i to it, where r0i – RV with distribu- tion function (d.f.) F0i(x); b) message from system S will transmit with probability i
)
0 t o( t pi
i ' '
P into outside medium and system’s income will decrease on value
0,
Ri where Ri0 – RV with d.f. Fi0(x); c) message from system Sj will transmit with probability Pjpji'to( t' ) into system S and income of system i S will in-i crease on value rji and income of system Sj will decrease on this value,
, , , 1n j i
j z where rji – RV with d.f. F1ji(x); d) message from system S willi transmit with probability Pipij'to( t' ) into system S and income of system j Sj
will increase on value R and income of system ij S will decrease on this, where i Rij – RV with d.f. F2ij(x), i, j 1,n; e) change of system S state won’t occur with i probability 1
Op0i Pi'to('t) at the time interval 't.It’s evident that rji Rji with probability 1, i,j 1,n, i.e.
) ( )
( 2
1 x F x
Fij ij , i,j 1,n (1)
Besides system S increases its income on valuei ri' during every small time inter-t val 't due to interest on money that are in it; let us suppose that r is RV with d.f. i
. , 1 ), (x i n
Fi We will suppose also that RV rji, Rij, r0i, Ri0, r are independent in i pairs, i,j 1,n.
Let dil( t' is income change of) i-th QS at the l-th time interval of length ǻt. Then from all that was saying before it follows
°°
°
¯
°°
°
®
'
'
'
z '
' '
z '
' '
'
' '
'
' '
'
).
( 1
y probabilit with
, , , 1 ), ( y
probabilit with
, , , 1 ), ( y
probabilit with
), ( y
probabilit with
), ( y
probabilit with
) (
0 0 0
0 0
t o t p
t r
i j n j t o t p t
r R
i j n j t o t p t
r r
t o t p t
r R
t o t p t
r r
t d
i i i
ij i i
ij
ji j i
ji
i i i
i
i i
i
il
P O
P P
P O
(2)
Income of system Si equals
¦
'm
l il i
i t v d t
V
1
0 ( )
)
( (3)
Let us introduce definition for expectation values (e.v.) correspondently:
^ `
rji xdFji x ajiM f
³
0
1 ( ) ,
^ `
,0
2ij( ) ij
ij xdF x b
R
M
³
f M^ `
ri f³
xdFi x ɫi0
) (
^ `
ri xdFi x a iM 0
0 0
0 f
³
( ) ,^ `
00 0
0 i ( ) i
i xdF x b
R
M
³
f , i,j 1,n (4)in view of (1)
ji
ji b
a , i,j 1,n (5)
Let us derive expression for e.v. of income of systemSi at the moment t.
According to formula for e.v. of discrete RQ
^
( )`
( )0 1
0
0 p a p b p c t o t
a t
d
M i
n
j ij ij i n
j
ji ji j i
i
il ¸¸' '
¹
·
¨¨
©
§
' O
¦
P P¦
, i 1,n (6)As it follows from (3) taking into account mǻt whent, 'to0
^ `
V t v c a p a p b p tM t v
n
j ij ij i n
j
ji ji j i
i i i i
i ««
¬ ª
»»
¼
º
¦ ¦
0 1
0 0
) 0
( )
( O P P (7)
For expected income for whole network considering (5), we have:
^ ` ¦ ¦
n>
@
i
i i i i i i i n
i
i t v c a p b p t
v t
W M
1
0 0 0
0 0
1
) ( )
( O P (8)
Let us notice that common expected network’s income doesn’t depend on ,
, 1 , ,
,R i j n
rij ij as that incomes absorb each other (if message transmits from one network’s system to another then income of the last one increases on some value and income of the first QS decreases on the same value).
2. Variations of network systems incomes
From expressions (2), (3) it follows that
^ `
2 21 0 0 0
0 0
2 V(t) v c a p b p a p b p t
M
n
j
ij ij i ji ji j i
i i i i i i
i ««
¬ ª
»»
¼
º
O P
¦
P PLet us introduce denotations:
^ `
ri a iM 02 20, M
^ `
Ri02 b2i0, M^ `
rji2 a2ji, M^ `
Rij2 b2ij, i, j 1,n (9)Let us consider expression
¨¨©§¦
' ¸¸¹· ¨¨©§¦
' ¸¸¹·21 2
0 1
0 2
0 ( ) ( )
) (
m
l il i
m
l il i
i
i t v M v d t v M d t
V M
d t d t i nM t
Md
m
l m
l j j
ij il m
l
il( ) ( ) ( ), 1,
1 1
1
2 '
¦¦
' '¦
z
(10)
Taking into account (2)-(9), we have:
^
d 't`
M>
r r r 'tr 't p 't@
M il2( ) 02i 2 0i i i2( )2 O 0i
> @ ¦ >
z
'
' '
'
n
i j j
i ji ji ij
i i
i
i R r t r t p t M r r r t
R M
1 2 2
2 0 0 2
0 2 ( ) P 2
'@
> ' ' ' @
'
¦
z n
i j j
ij i i
i ij ij ji
j
i t p t M R R r t r t p t
r
1
2 2 2
2
2( ) P 2 ( ) P
>
'@
''
Mri2( t)21 Op0i Pi t o( t)
( )
1
2 2
0 0 2 0
20 p b p a p b p t o t
a
n
i j j
ij ij i ji ji j i
i i i
i ' '
»»
»
¼ º
««
«
¬ ª
¦
z
P P
P
O , i 1,n (11)
Besides from independence of dil( t' and) dij( tǻ), jz considering (2) it fol-i, lows that M
dil(ǻt)diq(ǻt) o(ǻt)2. So as it’s fallowed from (10), (11) andǻt t,
m for variation of income of i-th QS when 'to0 we obtain the following expression
( ) 0^
( ) 02`
2^
( ) 0`
)
( i i i i i i
i t DV t v M V t v M V t v
DV
««
«
¬ ª
»»
»
¼ º
¦
z
t p b p a p
b p
a
n
i j j
ij ij i ji ji j i
i i i i
1
2 2
0 0 2 0
20 P P P
O
2
1 0
0 0
0p a p c a p a p t
b
n
j
ji ji j ij ij i i
i i i i
« i
«¬ ª
»»
¼
º
P O
¦
P P , i 1,n (12)3. Expected incomes of Jackson HM-network systems
Let us consider now the case when service rate of messages Pi(ki) in system Si
depends on number of messages that are present in it, i 1 n, . So we take off limita- tions that network must functioning in the term of high capacity. In our case expres- sion (2) will look as
°°
°
¯
°°
°
®
'
'
'
'
' '
'
' '
'
' '
'
' '
'
), ( ))
( ( )) ( ( 1
y probabilit with
), ( ))
( ( )) ( ( y probabilit with
), ( ))
( ( )) ( ( y probabilit with
), ( ))
( ( )) ( ( y probabilit with
), ( y
probabilit with
) (
0
0 0
0 0
t o t l k u l k p
t r
t o t p l k u l k t
r R
t o t p l k u l k t
r r
t o t p l k u l k t
r R
t o t p t
r r
t d
i i
i i i
ij i i
i i
ij
ji j j
j i
ji
i i i
i i
i
i i
i
il
P O
P P
P O
i j n
j 1, , z (13)
where ki(l) is number of messages in the i-th QS at the l-th time interval,
¯®
d
! 0 , 0
, 0 , ) 1
( x
x x
u - Heavyside function.
Let us find expression for expected income of system S at the moment t. Wheni realization of process k(t) is fixed, according to (13) and denotation (5), we can write:
^ `
««
¬
ª
'
¦
nj
j j
j ji ji i
i
il t k t a p a p k l u k l
d M
1 0
0 ( ( )) ( ( ))
) ( / )
( O P
) ( ))
( ( )) ( (
0
t o t c t p b l k u l
k i
n
j ij ij i
i
i ' ' '
»»
¼
P
¦
º , i 1,n (14)Then taking into account m't t and (3) we obtain
^
V t k t` ¦
M^
d 't k t`
a p c tM i i i
m
l
il
i 0 0
1
) ( / ) ( )
( / )
( O
'
¦
a p¦
m k l u k l tl
j j
j n
j
ji ji
1 1
)) ( ( )) ( P (
) ( )) ( ( )) ( (
1 0
t o t l k u l k p
b
m
l
i i i n
j ij
ij ' '
¦ ¦
P , i 1,n (15)When mof, 'to0
¦
m ' ' oto³
t j j jl
j j
j k l u k l t k s u k s ds
0 0 1
)) ( ( )) ( ( ))
( ( )) (
( P
P , j 1,n(16)
so
^
V t k t`
v c a p t¦
a p³
k s u k s dsM
t
j j
j n
j
ji ji i
i i i i
1 0 0 0
0 ( ( )) ( ( ))
) ( / )
( O P
³ ¦
n
j ij ij t
i i
i k s u k s ds b p
0 0
)) ( ( )) (
P ( , i 1,n (17)
Making average by k(t) and taking into account condition of normalization
( ) 1¦
kk t k
P for expected income of system S we will havei
^ `
¦ ^ `
k
i i
i
i t M V t v P k t k M V t k t
v ( ) ( ) 0 ( ) ( )/ ( )
««
¬
ª
c a p t
¦
P k t k¦
a p³
k s u k s dsv
t
j j
j n
j
ji ji k
i i i i
1 0 0
0
0 O ( ) P ( ( )) ( ( ))
»»
¼
³
t i ki s u ki s ds¦
jn0bijpijº0
)) ( ( )) (
P ( , i 1,n (18)
Let system S consists of i m identical service lines, the service rate of messages i in each line equals Pi, i 1 n, . In this case
), ), ( , min(
) ( ,
, ) ( ), )) (
( ( )) (
( i i i
i i
i i
i i
i i i
i
i k s m
m s k m
m s k s s k
k u s
k P
P P P
¯®
!
d i 1,n (19)
Then from (18) it follows
^ `
¦
uk i i i i i
i t M V t v c a p t Pk t k
v ( ) ( ) 0 O 0 0 ( )
»»
¼ º
««
¬
ª
u
¦ ³ ³ ¦
nj ij ij t
i i i
t
j j n
j
ji ji
ja p k s m ds k s m ds b p
0 0 1 0
) ), ( min(
) ), (
min( P
P , i 1,n (20)
It is natural to suppose that average of expression min(kj(s),mj) gives )
), (
min(Nj s mj , i.e.
) ), ( min(
) ), (
min(kj s mj Nj s mj
M , i 1,n (21)
where Nj(s) M
^ `
kj(s) is average number of messages (waiting and servicing) in system S at time intervali>
t0,t0s@
, i 1 n, . Therefore we receive from (20) the following relationc a p t
¦
a p³
N s m dsv t v
t
j j n
j
ji ji j i
i i i i
1 0 0 0
0 min( ( ), )
)
( O P
³ ¦
n
j ij ij t
i i
i N s m ds b p
0 0
) ), (
P min( , i 1,n (22)
As
³ ¦
¦
¦ ¦ ³
nj ij ij t
i i n
i i n
i
t
j j n
j
ji ji
ja p N s m ds N s m ds b p
0 1 1
1 1 0
) ), ( min(
) ), (
min( P
P
so expected income of whole network equals
^ `
W t¦
in ««¬ªvi ci a ip it ibi pi³
t Ni s mids»»¼ºM
1 0
0 0 0
0
0 min ( ),
)
( O P
For finding of Ni(t), i 1 n, , it can be applied recurrence by the time moments me- thod of analysis of average values for open QN [10].
4. Variations of incomes in Jackson HM-network
Taking into account (3), expression (22) can be rewritten as
c a p t b p
³
N s m dsv t v
t
i i i
i i i i i i i
0 0 0 0
0
0 min( ( ), )
)
( O P
»»
¼ º
««
¬
ª
¦
n j ji ji³
t j j i ij ij³
t i ij
ds m s N p
a ds m s N p
a
0 1 0
) ), ( min(
) ), (
min( P
P , i 1,n (23)
From (13), (19), (9) it follows (similar as we found (11)):