DSpace at VNU: Evaluation of CPM - A commit protocol for Mobile Distributed database systems

V N U JO U R N A L OF S C IE N C E . Nat ■ Sci . & Tech . ĩ XIX, N.j4. 2QQ3______________________ ___________________

EVALUATIƠN OF CPM - A COMMIT PROTOCOL
FOK M O B IL E D IS T R IB Ư T E D DATABASE SY S TEM S
Le H u u L a p
pnsts T e lecom m un ica tions In s titu te ()f Technology
N g u v e n K h a c L ic h , T r a n T h e T r u y e n
Research In s tĩtu tv <)f Posts a n d T rlc c o m m u n ic a tio n s
1.

I n t r o d u r tio n

Recent

e xp lo sio n

o f m o b iỉc

c o m p u tin g

has

p ro m o te d

th i*

research

and

d e vo lo p m e n t on m o b ile tra n s a c ỉio n s . e sp e c ia lly on d is tr ib u tc d da tabase system s

(M I)D B S ) |2, .‘ỉ, 8. 10). The M D D IỈS are ty p ic a lly b u ilt on top o f w ire d -c u m -w ire le s s
n rtv vo rk. w h i(‘h co n sists o f m obile devices (such as PDA, ia p to p , c*tc) ca lle d M o b ilc
l ĩn it s (M U ) c o m m u n ie a tin g w ith Fixed n e tw o rk v ia Base S ta tìo n s (HS) o r M o b ile
S u p p o rt S tơ tio n s (M S S ) (F ig u re 1) Each BS has n u m b e r o f w ire le s s c o m m u n ic a tio n
ehannels fo r M U s,
tra n s íic tio n

w h ic h

gencT ally

move and

in vo lve s

h a n d o ff fro m

p a rtic ip a tio n

ce ll to c e ll.

<>f seve ral

site s,

A d is trib u te d
w h ic h

u s u a lly

ilis t.rib u to ỚVer th e fix e d netvvork. b u t some o f the m may be in Nvirtíless n e tw o rk . For
the purpose o f tra n s a c tio n stu d y, th e netvvork can be vievved as a log ica l mess
( K ig u re 2).
Such M I)D B S m u st execute c o m m it protocols to e n su re a to m ic ity a fte r the

traiìsaction executions on data. Typically the commit protocols are based on
m rssage pa ssin# betvveen p n rtic ip n n ts in the tra n s a c tio n , in one o r m ore d iffe re n t
phases such as 1PC (O ne-Phase O o m m it), 2P C (Tw o-P hase C o m m it) and 3PC
(T hree-P hase C o m m it) |4. 5, 10). U n lik e w ith fíxed n e tw o rk f those protocols may
nnt w o rk w e ll in vvireless n ie d iu m and w ith the r ie x ib ility o f User m o h ility fo r
v a rio u s reasons |8, 10j:
•

w ire le s s

co n n e ctio n

is

prone

to

ía ilu re

due

to

the

n a tu rc

o f ra d io

p ro p a g a tỉo n lik e fa d in g , obstacle. and in te ríe re n c e ,
•

M U has lim ite d b a tte ry pow er, Processing speed. and s ta b le storage,

•

h a n d o ff ra te is u n p re d ic ta b le ,

•

lim ite d w ire le s s ch a n n e ls fo r s h a rin g betw een M U s,

•

ty p ic a lly t h i‘ c o m m u n ic a tio n d e lay is h ig h e r due to u n re lia b le physical
con nectio ns, and

•

fo r da tabase P a rtic ip a tin g sites are n o t u s u a llv in th e sam e n e tw o rk
p a rtitin n , th e y are co n n e ctin g fo r a sho rt đ u ra tio n .
35

Le H u u L a p , N guyên K h a c I J c h 9 T r a n The Truyen
Ilì th is p a p e r we e v a lu a te a new co m m it protocol • C P M (C o m m it P rotocol fo r
M o b ile ) th a t is do si^ne d to ove r come some lim ita tio n (>f th e m ost vvrdelv used 2PC
and to a llo w u se r-e n a h le d d isco n n e ctio n s and o fflin e Processing. S ection 2 provides
a sh o rt re v ie w o f s e v rra l c o m m it protocols and p o in ts out d ra w b a c k s o f 2PC in
m o bile c o m p u tin g . S ection 3 describes and analyzes th o th e o re tỉc a l aspects o f th e
C F M . T h e n e x t tw o sections p re se n t th e m o d e lin g and s im u la tio n o f th e C P M .

Figure 2: Logical view o f DDBS

Kigure 1: Generic physical model o f MDDBS

2. C o m m o p itm en t in d is trib u te d transaction Processing
2.1. T r a n s a c tio n M o d e ls
T h e re have been m a ny tra n s a c tio n m odels proposed fo r m o b ile c o m p u tin g [2,
3, 12] th a t c a p tu re cla ssica l tra n s a c tio n m anagem ent íe a tu re s such as co n currency
c o n tro l an d re co ve ry |1Ị and m o b ility . For the purpose o f th is p a per, th e y are
assum ed to a lre a d y e x is t. A d is tr ib u te d tra n s a c tio n T

in itia te d a t a s ite con sists o f

n u m b e r o f íra g m e n ts {T tJ} w h ic h a re d is trib u te d to p a r tic ip a tin g s ite s to execute.
The g e n e ric tra n s a c tio n

m odel considered

a p p lic a tio n s (A P P k/ fo r a r b itr a r v k e

in

th is

p a p e r c o n s is ts o f a set o f

J} o rig in a tin g tr a n s a c tio n T, a t M U s. A T,

re q u ire s th e in v o lv e m e n t o f a set o f p a rtic ip a n ts Ị ỉ >,/j € J, J =

a n d a set o f

c o rre s p o n d in g tra n s a c tio n c o n tro lle r ÍTC JỊ. Each tra n s a c tio n c o n tro lle r T C J handles
c om m an ds re la te d to tra n s a c tio n on b e h a lí o f p a rtic ip a n t p ’. O ne o f th e TC' is
chosen to be th e c o o rd in a to r (CO) fo r c o o rd in a tin g th e tra n s a c tio n in M a s te r/ Slave
ía s h io n . T h e re art* no rc q u ire m e n ts on th e lo ca tio n o f T O in th e netvvork and w itli
respect to

b u t th e c o o rd in a to r s h o u ỉd ho located in re lia b le w ire d n e tw o rk to

e n su re s a tis fa c to rv p e rform an ce.

2.2* C om m it p ro to co ls
It

is a s tric t

re q u ire m e n t o f a to m ic ity in database tra n s a c tio n s , th a t a

tra n s a c tio n T m ust e ith e r fu lly c o m m itte d o r abortecl at a ll p a r tic ip a tin g sites. To
e n su re th is p ro p e rty . a c o m m it p ro to c o l is exeeuted hy the tra n s a c tio n c o o rd in a to r,

E ỉ u Ị iK ítio n ()f ( 'PM

a c o m rn it p r o to e o l fo r ..

;ifte r a ll > itfs in lu rm i h r i r c n m p lc tio n o fT ,. A m o n g th e niost W1 (||*I\ rlcplo yed IS thi*
Ttvo-Phơsc C o n m iit r J lV ) p ro tn m l. w h ií h \ve w ill d e s c rib r in m ore cit»f;ii 1 in thí* next

su h srcĩion. Kvrn thuimlì. 2IH* IS ro n sid e rc d to be i n s u íĩic ir n t b c r a u s e C)f high
c o m m it ro st «1 1 1 ( 1 p o ír u t lỉil (>f 1/0 h ỉo c k in g . T h e re have becn m a n y a tte m p ts to
irn p ro v r th e 2 IH ' and to proposi* a ltc rn a tiv e s to solve these issm \s. Som c a im at
m in im i/m t' n is ts oỉ ||>H w r itin g and C om m unications, in w h ir h P rC (P re sum e d
C o m m it) and K a r lv P rcp a rc |w. 1 ]| an* o f thoso noticeable. O th e r re search in c lu đ e s
1IH 1 and v a ria n ts such iis ( 'L P (C o o rđ in a to r Log P rotocol), and ỈY V P ( Im p lic it YesVoti* P rotocol). 1PC Itf!iurt\s th e P R E P A R K phase o f ‘2 PC b u t its a s s u m p tio n s are too
s tr ic t to be tis H u l in ro m m e rria l svstem s. ỉn c o n tra s t to those ap pro ache s, 3P C
(th re e -p h a se c o m n n t) in tro đ u c e s m ore cost and c o m p le x ity bv an a d d itio n a l phase to
h c lp avoul th e I/O h lo c k in ^ in 2PC.

2.3. T ĩvo p h a s e c o m m i t p r o to c o l
Thi* ro m m it p ro to rn l. as 1 1> nam o in d ic a ỉc s , has tw o phases: th í' f ir s t asks fo r
v o tin g fro m a ll s ite s roíKỈy to c o m m it. and the second c a rry out th e c o m m itm e n t.
P h a s e 1: c o íu lils record « P K K I \ \ R K T , » to the log, w r itc s th e lug to s ta b le
storage and s ta rts a tim e r (F K K P A R E _ T IM E O U T ) th e n sends P R E P A R E message
to a l! T C . O n re e e iv in g such a message, each T ơ decides vvhether o r n o t to c o m m it.
l f the nnsvver is N O . T O ;ul(ls record « N 0 T „ » to th e log and responses A B O R T
T to CO. O th e rw is e . T í ' 1adds record « R E A D Y T „ » to the log a n d forces th e log to
the sta h le storage. T C ' re p lic s K E A D Y T „ to c o .
P h a s e 2 U pon re c e iv in g R E A D Y T,, fro m a ll TC ' w ith in tim e o u t p e rio d , c o
l*orce-\vrites T, log to s ta b lc stnrage. A t th is p o in t, th e fat(» o f th e tra n s a c tio n is
ilu te rm in c d . c o th e n simh I s C O M M IT T, to a ll T C 1. In th e bad ca se, th e P R E P A R E

phase tim e o u ts , c o logs T, and sends A B O R T T, to a ll T O . VVhen T C ' receives a
C O M M IT or A B O R T fro m c o , it vvrites the message to th e lơg, th e n ackn ow led ges
back to CO.
D espite th e p o te n tia l o f b lo c k in g , the tvvo-phase c o m m it p ro to c o l is c u r r e n tlv
thi* m ost vvidely ust‘d in p ra c tire fo r (lis trib u te đ databasc System s |1 0|. Hovvever, it
m ay be (liH ic u lt to a p p ly in m o b ile d is trib u te d e n v iro n m e n t.

For e x a m p le , the

m obile users m ay te m p o rn rily clisconncct to process its tra n s a c tio n fra tfm e n t o fflin e
fo r sa vin g povver, and th is actio n can lead to tra n s a c tio n a b o rt, r e s u ltin g in th e h ig h
abort fre que ncy. T h is is because o f a c ritic a l p o in t in 2PC, th a t a fte r tra n s a c tio n
executions fin is h , t h i‘ (lecision to c o m m it re lie s e n tire ly on th e chosen c o o rd in a to r,
w h ich may not hv iiNvarc oí’ m obile User preỉerence. Once th e (.’ () decides to c o m m it.
the connectio ns need to biĩ kept to avoid possible a b o rt due to tim e o u t* T h is can be
done rt*lia b ly in h ig h c Ịiia lity w ire d n e tw o rk , b u t th e re are no g u a ra n te e s in the
vvireless.


Le H u u L a p y N guyên K h a c L i c h y T r a n The T ru ye n
In a d d itio n , 2PC m ay n o t be s u ita b le fo r M U s w ith s m a ll m e m o ry because
the y have to m a in ta in a locai log to ha ndle CO*s dem ands a t a n y tim e . A n o th e r
weakness o f 2PC is due to its c o m m it n a tu re o f 2 phases r e q u ir in g 4p messages
(vvhere p

is

n u m b e r o f p a rtic ip a n ts ),

vvhich

are

ra th e r

exp e n sive

in

m o bile

Communications in te rm (>f tim e (and possibly m oney since th e m o b ile cost is
u s u a lly d e te rm in e d on th e basis o f connection tim e ).

3.

CPM • Comniit p roto col for m obile tra n sa ctio n m o d els
3.1. T h e C P M C o m m ỉt P ro to c o l
In th e lig h t o f above discu ssio n, it is reasonable to m in im iz e th e need to

co m m u n ica te over w ire le s s lin k and the n u m b e r o f messages p a s s in g in th e netvvork
per tra n s a c tio n . We proposed a novel com m it protocol c a lle d C P M (C o m m it p rotocol
fo r A /obile T ra n s a c tio n s ) th a t avoids th e possible d isc o n n e c tio n s d u rin g th e co m m it
process.

In

a d d itio n

CPM

enables

u se r-p ro a ctive

đ is c o n n e c tio n

an d

o fflin e

Processing fo r a r b itr a r y a m o u n t u f tim e . The d isco n n e ctio n m ay be in itia te d by
M U ’s tra n s a c tio n c o n tro lle r before decision to c o m m it is made.
O u r c o m m it p rotoco l C P M operates u n d e r ge neric m odel p re s e n te d in S ection
2.1. A lth o u g h each T C J (loes n o t need to be a t the same s ite as c o rre s p o n d in g P \ T C k
is set a t M U site to a vo id possible c o m p le x ity in local c o m m u n ic a tio n v ia vvireless
m e dium . In C PM , th e tra n s a c tio n c o n tro lle r o f M U (T C k) w h e re a tra n s a c tio n T, is
o rig in a te d plays th e ro le o f c o o rd in a to r in 2PC d u rin g tra n s a c tio n e xe cu tio n . Ít
d ivid e s th e tra n s a c tio n T, in to ừ a g m e n t set
\v rite s

th e

tra n s a c tio n

e x tra c ts its íra g m e n t T lk and force-

log to sta b le storage.

In s te a d

o f s e n d in g

th e

re st o f

tra n s a c tio n T ầị - T ,k to c o o rd in a to r as in 2PC, T C k d is tr ib u te s those fra g m e n ts to
co rre sp o n d in g p a rtic ip a n ts and s ta rts a T R A N S _ E X E C tim e o u t. Each p a rtic ip a tin g
site executes its p a rt a n d send co m p le tio n message back to th e re q u e s te r.
I f th e T C k receives a ll such messages, it s ta rts th e c o m m it process by sen ding
C O M M IT _R E Q U .E S T message and tra n s a c tio n log to c o . U po n re c e iv in g c o m m it
re que st and tra n s a c tio n log fro m T C k, c o fo re -w rite s th e log to s ta b le storage and
sends C O M M IT message to a ll p a rtic ip a tin g sites. Each s ite th e n c o m m its its
tra n s a c tio n

p o rtio n

T M and

acks

back

to

co

acknow led gem e nts are received a t c o , th e c o

ònce

fin is h e d .

If

a ll

such

logs th e tra n s a c tio n and sends

C O M M IT _ A C K message back to T C k. T h e tra n s a c tio n is c o m p le te d fro m now on and
CO can íb rg e t about it. N o te th a t th e re is no need to send C O M M IT to T C k because
th e C O M M IT .A C K is a lw a y s em bedded in C O M M IT .R E Q U E S T b e fore ha nd. T h u s
the n u m b e r o f messages p a ssin g vvireless m e dium is m in im iz e d .
O th e rw is e

th e

TR AN S_EXEC

tim e o u t

is

trig g e re d

be fo re

T C k gets

a ll

T R A N S .A C K fro m p j, T C k sends A B O R T message to c o and T R A N S _ N O T _ O K to
A P P k. In th is case, c o has n o t logged the tra n s a c tio n y e t. so th e A B O R T message


E v a ỉ u a t i o n ()f C 'ỉ * \ i

a v o m m it p r o to c o l fo r .

m>i‘ds to c a rry a ll n o crssa rv in fo rm ;itio n ab o u t th e ab o rte d tra n s a e tio n (i.e. ID o f
p a r tin p iin ts

U s in ^ g iv rn in fo rm a tio n , c o m u ltic a s ts A B O R T messages to a ll

TC ' {J * h) T ( ’ t h r iì t r lls its co rre s p o n d in g p tơ a b o rt.
As m e iìtio n e d abọvo, tho p a rtic ip a n t P' and the a g ent T ơ a c tin g on its b e h a lf
a r r not n e ce ssa rily at t h i’ sanu' s it t \ so d u rin g the c o m m it process, p* m ay fa il and
m ay Iio t v e n tv th e c-ommit w ith T O . So T C J needs to use a C ()M M !T _ K X E C tim c o u t
vvhriì c o m m a n d in g r

Upon re c e iv in g such com m and, p rin is h e s its tra n s a c tio n

p o rtio n a n d th e n lo tf-w rite s a “ c o m m it” . The tra n s a c tio n p o rtio n is done. I f the
lim e o u t is passed and P' has not c o n firm e d th e c o m m it, T C 1 assum es th e ĩa ilu re o f
l >!. T C th e n checks the existonce o f “ c o m m it'’ log a t P \ I f th e log cỉoes no t e x is t, the
T C ’ tu ll p to rtj d() th i' tra n s a rtio n e xe cution and c o m m it prơcedures.
Once re c e iv in g C ()M M IT _ A C K fro m c o . T C k sends T R A N S .O K to A P P k. the

vvhole tra n s a c tio n is ỉu lly com pleted. The C PM can be expressed s e m i-fo rm a lly in
te rm o f pseudo code as in P igure 3 -F ig u re 6.

A c ỉ i v i l i c s uỉ A P P :
b e g in
send

tra n s a c tio n

w a it

u rvtil

i f ( t he

e ls e

T

•

t- 0

.

TC

k

•

re c é iv in g

.

r-

in s tru c tio n

rr or ĩ ì TC

i n s t r u c t . i o n . i s TRANS OK)
s t a r t new t r a n s a c t i o n a f t e r

//th e

t r a n s a c t io n

restart

a

tim e

in te rv a l

25 a b o rte d

t.ra n s a c tio n

afte r

a

tim e

in te rv a l

end
F ig u re 3: CPM a lg o rith m at a p p lỉca tio n APP

/\( 7 / vilics cit T C :
B e g in
k
if(T C

re c e iv e

b e g in
k
TC

send

a

tra n s a c tio n

T.

to

TC" w r it e

t:he

a 11

TCk è x t r a c t

s c h ed u ler

«LOG

its

//th is

T ;)

T, »

to

and

p la y s
then

th e

ro le

of

g e t írag m en ts

RAM

p o rtio n

T lk

ị T.

T lk}

Ỳ

T C ' send

the

TC

process

TC

st.art.

rest
its

tim e r

i f ( t im eout

-

p o rtio n

not

a ll

co rre s p o n d in g

fo r c o n firm a tio n

re c e iv e

a ll

send

ABORT T

TCk

senđ

TRANS

to

co

NCT_OK T ,

co A P P k

end
e ls e

TC

ro -e iv e

311

TRANS ACK:

from { p ' }

T RANS_ACK

b e g in
TCk

p}

T lk

and w a i t

&Ẳ TC

to

of

T )

TC

k

{Tt l|


Le H u u L a p , N g u y ê n K h a c L ic h , T r a n The T ru y e n

40

b eg in
T C k s e n d COMMI T T ; , a l l < and i t s ack o f fra g m e n t
to co
TCk w a i t f o r r e s p o n s e f r o m c o

i f ( T C k r e c e i v e COMMI T_ACK T i f r o m
TCk send

end
and
/ / F r o m now o n TC p l a y s
e ls e

if(T C '

re c e iv e

th e

T RA NS_ 0K

r c le

COMMI T T i

to

of

from

CO)

APPk

TC
CO)

b e g in
TC

send

record

«C O M M IT

T .,»

and

c o mm an d C O M M I T
p-

T,

to

p

TC s t a r t t i m e r a n d w a i t f o r a c k o f T j , f r o m
i f (TC
r e c e i v e com m it c o n f i r m a t i o n o f p )
TC1 send
e ls e

C OMMI T_A CK T j j

to

co

/ /tim e o u t
b eg in
//p ;

fa ils

TC
c h e c k l o g a t pi f ( e x i s t t h e l o g < < C O M M I T T 43» )
s e n d COMMI T_ ACK T, . t o c o
e ls e
TC: ask

p3 t o

redo

the

tra n s a c tio n

p o rtio n

based o n l o g f r o m c o

end
end

e ls e

/ / T C j r e c e ì v e ABORT Tịj f r o m c o
send ABORT T n t o p j

end
F ig ure 4: C PM a lg o rith m a t tra n sa ctio n c o n tro lle rs (T C k & T C 3)
A c tiv itie s a t P :
b e g in
if( p
r e c e iv e
b e g in
execute
send
wai t

tra n s a c tio n

fragm ent

T,

from

TC )

Tn

T R AN S_A CK T n b a c k t o TC*
f o r com m it fro m c o r r e s p o n d m g

TC:

end
e ls e

r-e c e iv e C OMMI T T. fro m TC )
b e g in
com m it th e p o r t io n Tn
f o r e - w r i t e l o g <<COMMIT T M >> t o s t a b l e
end
e ls e i f ( p r e c e i v e ABORT
f r o m T C 1)
a b o r t i t s p o r t i o n T í:
end
i f (p

storage

F ig u re 5: C P M a lg o rith m a t p a rtic ip a n t 1J

T jj


K v a lu ơ tio n o f CPM

(t c o m m it p r o to c o l fo r ..

•11

À c t i v i í i c s (lí ( ( l i
b e g in
i f ( 0 ’ re^eive ABORT T
i f ( r.

<

t r■■:m TC*)

ĩ. 'G T . • * »?x i s t _ )

• • x t r i ■’

: i : ‘ 4' i p a n t

IL

fr*r.

message

Ab OR T

T.

e ls e

e ls e

e x a m i n e <<L.OG T ; > > t o g e t
s e n d ABORT Tj t o a l l T C ’
i f (CO r e c e iv e a l l < < L OG !![,>>,

TC1

COMMIT T. and COMMIT ACK T., from T O
b e g in
f o r c e - w n t e <<!,.'■; T >> to stabỉe storage
send COMMIT T to a 11 TC'
w a i t f o r a c k s f r o n TC
if(CO

re c e iv e

a ii

c o m p iete
send

the

C CM MI T _ AC K T
c ra n s a c tio n

COMMI T ACK T,

to

from
T,

TC )

and

T C'

end
end
F ig u re 6: CPM a lg o rith m at co o rd in a to r c o
3.2. C ost a n a ly s ỉs <)f C P \ Ỉ
The CPM is designed w ith the cost m in im iz a tio n in m ind , as show n in Table 1.
T a b le 1: Cost com parison o f co m m it protocols
P r o to c o l

N u m b e r o f m essage s

L o g fo rc c w r ite s

C o m m u n ic a tio n
d e la y

2PC

4 (n -l)

l+ 2 n

4

PrC

3 (n -l)

l+ 2 n

4

EP

n -1

2+n

2

CL

n-1

IY V

2 (n -l)

1

2

CPM

2 < n -l)

1+n

2

l+ (n + n_op)

2

w here n op is n u m b e r o f tra n s a c tio n ĩra g m e n ts .
A lth o u g h C F M a llo w s fle x ib le o fflin e P r o c e s s i n g and d is c o n n e c tio n d u rin g
c o m m ittin g , b u t whi*n too m anv tra n s a c tio n s w a itin g to in the c o o rd in a to r queue,
the c o o rd in a to r is forced to c u to ff so me tra n s a c tio n s to avoid process o v e rflo w . The
c u to ff m echanism can be as sim p le as ro u n d -ro b in to rem ove th e ơ ldest ite m in the
queue, b u t th is is open to la te r im p le m e n ta tio n s

4. S im u la tio n m odoling
In a d d itio n to q u a lita tiv e a n a ly s is presented in p rc v io u s sections. we have
q u a n tita tiv e ly in v e s tig a te d th í4 pc»rformance o f C PM in d iffe re n t scenarios and in


Le H ư u L a p t N g u y en K h a c Lichy T r a n The T ru y e n

42

co m p a riso n w ith th e 2 P ( \ Wi* doveloped a D iscrete K vent S im u la to r (I)E S ) th a t

runs the model of the generic wired*cum-wireless network topology (see Pigure 1).
T h e m a in idea b e h in d the DKS appro ach to s im u la tio n is based on con cept o f e ve nt,
an occurrenct? in thi* system at d is c re te tinu*. K vents are o fte n c re a te d in advance
ahead oí c u rre n t

ti nu* and scheduled and a c tiv a te d a t p re d e fin e d tim e -s ta m p

associated w ith th e m .
T h e m a in flo w u n it in th e system model is message (d a ta & s ig n a lin g ), th a t is
a p p ro p ria te fo r c u rre n t m odel <>f m o b ile C om m unications. T h e c o m m it p rotoco ls are
o p e ra tin g on th e a p p lic a tio n la v e r w here the u n đ e rlv in g n e tw o rk p ro vid e s the
tra n g p a re n t c o m m u n ic a tio n ch a nne ls. A lth o u g h o u r s im u la to r s u p p o rts a r b itr a r y
n e tw o rk

to p o lo g y.

but

fo r

the

purposes o f th is

paper,

th e

fix e d

n e tw o rk

is

im p le m e n te d as a ge neric node vvith p rim a ry fu n c tio n a lity o f ro u tin g . F o r s im p lic ity
we do not im p le m e n t a n y m echanism fo r che cking fo r e r ro r a n d message o rd e rin g a t
T ra n s p o rt la y e r as u su al in c o m m u n ic a tio n n e tw o rk , th e loss o f message and
d isco n n e ctio n a n d re co n n e ctio n in w ire le s s lin k are m odeled as s u ffic ie n tlv long
de lay. T h is is because it is u s u a l fo r tra n s p o rt la v e r to assum e message loss based
on tim e o u t w h e n n e tw o rk congestions occur. S im ila rly . h a n d o ff is also m odeled as
long đ e lay vvith th e average d e lay tim e fo r h a n đ o ff is ls . T a k in g in to account the
fact th a t ave rag e v e lo c ity o f m obile users is 3m/s, avcrage c e ll d ia m e te r is a ro u n d
lỉOOm. so th e h a n d o ff ra te is ab o u t one every lOOs.
Each s ittí co n sists o f an a p p lic a tio n APP, a tra n s a c tio n c o n tro lle r T C and a
da tabase (o r p a rtic ip a n t). T ra n s a c tio n s are alvvays in itia te d a t M U a n d the nevv One
is cre a te d

Is a fte r th e c o m p le tio n o r a b o rt o f the p re v io u s . Each p o rtio n T tJ o f

tra n s a c tio n T is a s s ig n e d ^ o a p a rtic ip a n t in the m *tw ork ra n d o m ly , t*xccpt fo r the
o r ig in a tin g si te (M U ), lỉecause th e re are m u ltip le tra n s a c tio n s b e in g c o n c u rre n tly
re q u e stcd a t c o and p a rtic ip a tin g site s, each s ite needs to m a in ta in a tra n s a c tio n
queue. F o r s im p lỉc itv . thí* qucue is im p lc m e n te d on K irs t C om e K irs t Serve (FC FS )
basis. To a v o id po ssible Processing overload. the tra n a a c tio n qu e u e a t c o operates
in ro u n d -ro b in s tra te g y , w here the o ld e s t a v v a itin ^ tra n s a c ỉÌH iì w ill be a b o rte d when
th t‘ queue is fu ll.
A t each s ite , C PU e xccu tio n an d 1/0 ciisk access are im p le m e n te d in m u tu a l
e x c lu sive

fa s h io n

ta k in g

a c e rta in

am ount

<)f timc* d e p e n d in g

on

n a tu re

of

o p e ra tio n s. O th e r re que sts have to w a it u n til thc» c o m p le tio n o f the previous.
C u r r e n tly no p a ra lle l P rocessing is m odeled, b u t it can he s im u la te d hy a d ju s tin g
th e average P rocessing tim e .
We use th e mean d e la y ()f lO m s (V. K u m a r. 2002) w h e n s e n d in g message in
w ire le s s e n v iro n m e n t. and in w ire d n e tw o rk . th e average tra n s m is s io n tim e is 5ms.
P a ra m e te rs used in s im u la tio n m odels are shown in T a b le 2


E v a l u a t i o n o f C P M - a c o m m it p r o to c o l fo r ...

43

T a b lc 2: l\ir;im <*t(*rs s e ttin g in s im u la tio n

ỈAỉự, Kon r \VTIU*

200ms

T r;m s:ictio n S r^ iìirn t K xiTU tion

34ms

T i nu* to Si'í om* lock

lm s

T im e to unlock

1ms

l) ;it;i ()hji*( t updati* t ime

6ms

T ra n sa ctio n Díĩgrec o f (iis trih u tio n

7-10 íra g m e n ts

Messa^e h ỉin d lin ^ tim e

lm s

Averagt* w ireless tra n sm issio n delay

lOms

Àvi»ratft‘ vviređ tra n sm issío n đelay

5ms

Database system degrue o f d is trib u tio n

>= 10 sites

T im e o u t

60s

T h o s im u la tio n s o ftw a re is developed in S t a n d a r d C ++ to p ro m o te p o r ta h ility .
C u rr e n tly it su p p o rts 2PC and CPM w ith s im ila r vvorking c o n d itio n s so th a t th e y
can be com pared w ith each o th e r.

5. E x p erim en ts and R esults
W t'

have

co n ducted

a

com pre hen sive

e x p e rim e n t

set

to

e v a lu a te

th e

p tT Íb rm a n a * o f tw o c o m m it protocols o f in te re s t: 2PC and C P M . T h e p e rfo rm a n c e is
p r im a r ilv m casured u s in g m e tric oí* tra n s a c tio n th ro u g h p u t pe r I in it o f tim e . O th e r
facto rs includ e p o rtio n o f successful tra nsactio riy average n u n ib e r o f messages p e r
tra n s a c tio n

and

trơ n s a c tio n

tu rn a ro u n d

tim e

(F ig u re

10).

The

tra n s a c tio n

tu rn a ro u n d tim e is (ie fin e d as th e d u ra tio n betw een the s ta r t and th e end o f the
successíul tra n s a c tio n a t o r ig in a tin g a p p lic a tio n (a t M U s in th is s tu d y ). T h e re s u lts
presented in th is p a p c r are o n ly those o f th e m ost in te re s t.
F o r the systen i a rc h ite c tu re s e ttin g ' each tra n s a c tio n T, can o p tio n a lly selects
a nevv c o , bu t fo r purposes of pe río rm a n ce s tu d y u n d e r he avy loads, th e re is o n iy
one CO is set fo r the w h o le s im u la tio n process, fo r both cases o f 2PC a n d C P M . For
most stuclies, th o netvvork consists o f 10 DBS b u t the s im u la to r yllovvs a r b itr a r y
n u m b e r o f DBS.
5.1. P e rfo rm a n c e c o m p a r is o n u n d e r n o r m a l c o n d itio n
T h is e x p e rim e n t in v e s tig a te s th ro u g h p u t o f 2PC and C P M u n d e r load o f
m u ltip le p a ra lle l a ctive M ư s . The system ru n s u n d e r th e d is c o n n e c tio n p r o b a b ility
on w ireles8 lin k o f 0.5% • in n o rm a l s itu a tio n . The queue size is 10 se t fo r b o th 2PC

and CPM.
U n d e r th is c o n d itio n , th t‘ C P M p e rfo rm s b e tte r th a n 2PC u n t il i t reaches
m a x im u m (K igure 7). A í t r r th a t p o in t. th e system th ro u g h p u t ()f C P M decreases


Le H u u I x i p , N guyên K h a c Lichy T r a n The T r u y e n

44

ra p iđ ly w h ile the 2VC p e río rm a n ce degrades ra th e r g ra d u a lly . F o llo w in tf th e p o in t
w here n u m b e r o f a c tiv e M ư s is ro u g h ly 35, th e C P M w o rk s w orse th a n 2PC does.
F ig u re 9 shows th e average nu 111 be r o f message ()f th e tw o protoco ls. T h is re a ssu re s
p re vio u s th e o re tic a l a n a ly s is (T a b le 1). F ig u re 8 p re se n ts th e sam e s tu d y as in
Figure 7, b u t u n d e r vveak w ire le s s e n v iro n m e n t (d isco n n e ctio n ra te 4%). T h is v riìì be
s tu d ie d in thc» n e xt sub section.
m

350

F igure 7: Períbrm ance o f 2PC and C PM
under load vvịth norm al disconnection rate
(0.5%)

F ig u re 8: P eríorm ance o f 2PC and C P M
u n d c r load w ith lìi^ h disconnection ra te
(4%)
3500

20

76

50

Number of active MUs

40

60

Number of active paraftel active MUs

F ig u re 9: A veragc n u m b e r o f messages per
tra n sa ctio n

Average transaction turnaround time

10

ếOlivr Mư60

60

F ig ure 10: Average tra n s a c tio n tu rn a ro u n d
t i me

5.2. E ffe c t o f h a n d o ffs a n d iv ire le s s lin h f a ilu r e s
The com m on p ro p e rty betvveen hando ffs and w ire le s s lin k fa ilu re s are long
d elay (ra n g in g fro m seconds to h o u rs ) com pared to n o rm a l tra n s m is s io n tim e (1lOms). H ow eve r, the average h a n d o ff tim e u sin g in th is s tu d y is a ro u n d ls , whim uch less th a n ty p ic a l tim e o u t v a lu e (m in u te s to ho u rs). W y ru n th e s im u la tio n o f
v a rio u s average h a n d o ff ra te s ra n g in g fro m no h a n d o ffs to 2 h a n d o ffs /m in u te , or
w ith th e speed o f lO m /s (or 3 6 k m /h ) across the cell d ia m e te r o f 300m . The re s u lts
c o n ĩirm

our

in tu itio n

as

th e re

th ro u g h p u t as h a n d o ff ra te grows.

is

no

s ig n iíìc a n t

de gra de

in

o v e ra ll

system


E v a ỉu u tio n o f CPM

45

a c o m m it p r o to c o l fo r .

2PC under various disconnection

20

F ỉg u re I I : Kffect o f discunnertìon
íìvc|iu*ncy OM o ve rall th ro u g h p u t (handoíT
ra te is ().()])

NumbéKbf MUs

60

00

F ig u re 12: P erlorm ance o f ‘2 PC w ith
d iffe re n t tra n sa ctio n disconnection ratvs
(0.5%. 4%, 12%)

CPM unđer vartous disconnection probability
— ♦— CPM • d«s 0 5%

350

Number of MUs

F tịỊitrv 13: Pertbrmance o f C l\ \ l w ith diAerent transaction disconnection rates (0.5°'o, 4°o)
Thí* m*xt e x p e rim e n t (K ig u re 11) is to in v e s tig a te how íre q u e n t d isco nne ction
chan^es o v o ra ll system th ro u g h p u t. The đ isco nne ction ừ equency is th e p ro b a h ility
th a t a tra n s a c tio n can bi' disconnected (iu rin g its iife tim e . A lth o u g h th e p ro b a h ility
be ing stu d ie d is r a th e r a r tific ia l (up to 12%), th e s im u la tio n d e m o n s tra te s th a t both
2 PO and CPM p e ríb rm a n ce drops ra p id ly as th e p ro b a h ilitv o f vvireless lin k ía ilu re
increases. T h is is due to tim e o u t occurrences, d u rin g the c o m m it process in 2PC and
d u r in t' thí* tra n s a c tio n exe cu tio n in C P M .

In case o f 2PC. th e deeision to a b o rt due

to tim e o u t in P R E P A R E phase is m ade e n tire ly by c o . T h is a b o rt niessage m ay
t iik r s ig n iíìc a n t a m o u n t ()f t im r to reach T C k to schedule th e tra n s a c tio n re s ta rt. In
C I\\1 . fo r most o f th e tim e , th e tim e o u t occurs at T C \ and th e tra n s a c tio n re s ta rt is
im m e clia te ly scheduled. T o ^ e th e r w ith the fact th a t th e n u m h e r o f messages fo r
each successful tra n s a c tio n in 2PC is la g e r th a n th a t in C P M . one can expect th a t
CPM p e rfo rm s c o n s is te n tlv h e tte r on o v e ra ll.

T h e p e río rm a n ce ní ( T M u n d e r load vvith d iffe re n t ỉra n s a c tio n d isco nne ction
p ro b a h ility (K ig u re

13) tends to have the same c h a ra c te ris tic s as th a t o f 2PC


Le H u u L a p , N guyên K h a c ỈÁ chi T r a n The T ru y e n
(K ig u re 12). T h a t is thi* vveak c o m m u n ic a tio n m e d iu m m akes th e tim e to reach
m a xim u m th ro u g h p u t lo n g e r and also decreases th e m a x im u m v a lu e .
5.3. S ystem p e r fo r m a n c e in r e la t io n w it h q u e u e 8Ìze a t c o
It

is in tu itiv e

th a t

th e queue size a t c o

in flu e n c e s th e o v e ra ll system

p e ríbrm an ce because o f the ro u n d -ro b in o p e ra tio n . A set o f e x p e rim e n ts has been
conducted to c o n firm th is con clusion (F ig u re 14). T h e o v e ra ll th ro u g h p u t does no t
change w ith queue size w hen the n u m b e r o f p a ra lle l a c tiv e M U s 1 8 s m a ll. H ow eve r,
it appears to d e te rm in e w hen the system degrades u n d e r heavy load. T h e la rg e r
queue size, th e b e tte r it can h a n d le m u ltip le a w a itin g tra n s a c tio n s , vvhich is c le a rly
ro u g h ly

p ro p o rtio n a l

to

n u m b e r o f a ctive

Mưs

o p e ra tio n a l

in

the

n e tw o rk .

S im ila rly . th e average tra n s a c tio n tu rn a ro u n d tim e is d ir e c tly re la tữ d to queue size
in closed queue th e o ry (F ig u re lõ ).
?900

CPU Pertormance w ith v a n o u s q u e u e Síze

T umaround time

Queue 10
Queue 30

10

20

30

40

50

60

o f A C í-ve

Figure 15: Average transaction
F ig u re 14: Inílu en ce o f queue size a t c o on
System th ro u g h p u t

tu rn a ro u n d tim e vvith d iffe re n t queue
size (10,30.50)

5.4. Other experiments
A s ig n iíìc a n t n u m b e r o f o th e r e x p e rín ie n ts have been c a r ric d o u t in c lu d e th e
in v e s tig a tio n s o f v a rio u s s tra tc g ie s to s ta rt an re s ta rt a tra n s a c tio n , d iffe re n t
p ro b a b ilis tic d is tr ib u tio n s o f ra n d o m h a n d o ff and đ is c o n n e c tio n tim e , tim e o u t valu e
analyses and C P U speeds. Those re s u lts a ll e o n íirn i m a in c o n rlu s io n s presen ted in
th is paper, o f w h ic h we s u m m a riz e in n e xt Section.

6. C on clu sio n s and P urth er work
We havt* proposed a new c o m m it protocol fo r rnohile tra n s a c tio n
(C P M ). F ro m th e o re tic a l a n a ly s is and q u a n tita tiv e s tu d y

m odels

u s in g s im u la tio n , we

found th a t C P M p e rfo rm s c o n s is te n tly b e tte r th a n th e c la s s ic a l tw o-pha se c o m m it
(2PC)

protocol.

The

s im u la tio n

approach

has

proved

a

g re a t

aid

to

the


Evtt l u a t i o n o f C P M

a c o m m it p r o to c o l f o r ..

47

u n d c rs ta n c lin g o f protocol be havio rs and correctn ess u n d e r v a rio u s c o n fig u ra tio n s .
T h ỉ' n u nn re s u ỉts frorn e x p e rim e n ts presented in th is p a p e r are s u m m a riz e d below:
•

M ost ()f the ti me the CPM p e rfo rm s s ig n ific a n tly b e tte r th a n th e classical

2PC under mobile coníigurations.

In adciition, CPM allows offline

Processing and M U -enablecỉ (iisco n n e ctio n d u rin g the c o m n iit tim e .
•

H a n d o ff ratí* is not very c r itic a l in system pe río rm a n ce

•

Disconnections d u rin g t r a n s a c ti o n d e g ra d e t h e p e rí o r m a n c e rapidly.

•

T u rn a ro u n d

ti me tenđs

to

he

p ro p o rtio n a l

to

size

o f the

queue

at

c o o rd in a to r. T h is illu s tra te s th a t the q u e u in g m odel can he a p p lic a b le to
th is problt»m. It is in co n clu siv e ab o u t th is tim e in case o f 2PC and CPM .
•

T h e fo llo w in g extensions to th is research are in c o n s id e ra tio n :

•

U s in g

M o b ile

A gent

(M A )

fo r

tra n s a c tio n

m anagem ent

in

m obile

d is tr ib u te d database system s
•

T h e s o ftw a re vvill ho re -de sig ne d as a fra m e w o rk fo r s im u la te c o m m it
p ro to co ls and to enable th e p ro o í o f correctn ess u s in g m ore p o w e rfu l
te c h n iq u e s such as Colored P e tri nets [7ị.

R efer en ces
1.

I \ A . B e r n s t e i n , V. Had zi l ac o s, a n d N. G o o d m a n , C o n c u rre n c y c o n tro l a n d recovery in

d a ta b a s e s y s te m s , Adcỉ ison- Wesley, 1987.

2

V K. C hrysanthis, Transaction Processing in mobile com puting environm ent, Proc. IEEE
VVor k s h o p A d v a n c e s in P araỉlel a n d D is tr ib u te d S y s t e m s , Oct . 1 993, pp. 7 7- 8 2 ,

3.

M .H

D u n h a m . A. H e l a l , s. B a l a k r i s h n a n , A mo b i le t r a n s a c t i o n m o d e l t h a t c ap t ur e s

both the data and movement behaviour, A C M /B a ltz e r J. Specỉũl Topics in Mobỉle
Netivorks and Applications, 1997, 149-162.
ị

A. C r a y , A c o m p a r i s o n of t h e b y z a n t i n e a g r e e m e n t p r o b l e m a nd t h e t r n n s a c t i o n c o m m i t
problem, LNCS448 , Springer Verỉag. 1987.

5.

»J. C ra y, Notes on database opera ting systems, Lecture Notes in C om puter Science,
V oi. tiO S p rin g e r V erlag, 1978.

6.

K. G u p t a , J.

da ta ba se

Haritsa,

systems,

K. R a m a m r i t h a m , R e v i s i t i n g c o m m i t Processi ng in d i s t r i b u t e d

Proceedings o f ACM SIGMOD In te rn a tio n a l
Management o f Data, May 13-15, 1997, Tucson, Arizona, USA.

Conference

on

7.

L. M Kristensen, s. Christensen, and K. Jensen, The practitioner guide to coloured
p e t r i nets. I n t. J S 7 7 T ( 1 9 9 8 ) 2 : 98 132.

8

V Kumar, N. Prabhu. M. H Dunham, and A. Y. Seydim, TCOT - A Timeout-Based
mobile transaction commitment protocol, I E E E Transactions on C o m p u t e r s , Vol. 51, No.
10. O c t o b e r 2 0 0 2 .

9

(V Mohan, B, Lindsav and R. Orbermarck, Transaction management in the distributed
đ a t a b a s e m a n a g e m e n t System. A C M T r a n sa c tio n s on D a ta b a s e S y s t e m s , 1986.


Le H u u L a p y N g u y c n K h a c L i c h , T r a n The Truyen

48
10. A. S i b e r s c h a t z , H . F. K o r t h a n d

s.

Sudarshan,

Database systern concepts, 3rd ed. M c

G ravv-Hill. 1997.
11. J. Stamos and F. C ristian, A Low-Cost Atomic Commit Protocol, Proc. o f 9th synmp. on
reliable sistributed sỵstems, Oct 1990.
12. G.D.YValborn, P.K.Chrysanthis, Transaction Processing in PRO-MOTION, proceedings of
the ì 9 9 9 A C M S y n i p o s i u m on cipp hed c o m p u tỉn g , S A C '99, S a n Antonio, T X , USA, 1999.
pp. 389-398.

TAP C HÍ K H O A H O C D H Q G H N . KHTN & CN, T X IX , N04, 2003_________

Đ Á N H G IÁ G IA O T H Ứ C H ộ p T H Ứ C C P M
C H O C Á C H Ệ C ơ SỞ D ử L IỆ U P H Â N T Á N 1)1 Đ Ộ N G

L ê H ữu L ập
Học viện Công nghệ Bưu chính Viễn th ô n g
N g u y ể n K h ắ c L ịc h , T r ầ n T h ế T r u y ề n
Viện K hoa học K ỹ th u ậ t Bưu điện
Hệ th ố n g cờ sở dữ liệ u

ph ân tá n d i động (M D B S -

M o b ile D is trib u te d

D atabase S yste m s) gần đây th u h ú t các nhà nghiên cứu với n h iề u các mô h ìn h giao
tác khá c n h a u đà được để x u ấ t. Các giao thứ c hợp thứ c (C o m m it P roto col) cổ điển
như 2PC (Tvvo-Phase C o m m it) có kh á nh iều điểm yếu tro n g m ôi trường này do tín h
chấ t của th ò n g tin vô tu y ế n và sự d i động của người dùng. B à i báo này để x u ấ t một
giao thứ c hợp thứ c mới gọi là CPM (C o m m it Protocol fo r M o b ile ) có th ể hoạt động tố t

trê n các mô h ìn h giao tác khác nhau, tro n g đó cho phép xù lý ngoại tu y ế n và ngắt kết
nỏì tạ m thờ i tro n g k h i đang thực h iệ n m ột giao dịch dữ liệ u q u a m ạng d i động.
G iao th ứ c C P M có sô lượng bản tin , sô lân v iế t n h ậ t k ý và o bộ nhớ ổn đ ịn h và
chi p h í tru y ề n th ô n g ít hơn 2CP, th íc h hợp với m ôi trư ờ ng tru y ề n th ô n g kém tin cậy
như di động. C P M cho phép động ch ủ động cao hơn từ phía d i động th a y vì lệ thuộc
hoàn to à n vào bộ điề u phối tr u n g tâ m như với 2PC và các giao thứ c khác.
C h ú n g tô i xây đựng một phẩn mềm mô phỏng m ạng d i động theo hướng sự
k iệ n rời rạ c (D E S - D iscre te E ve n t S im u la tio n ) với giao thứ c hợp thứ c 2PC và CPM
ho ạt động trê n đỏ. Các kế t quá mô phòng đà chứng tỏ dược h iệ u s u ấ t cùa C P M cao
hơn 2PC tro n g m ôi trư ờ ng phân tá n d i động.