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

EVALUATIƠN OF CPM - A COMMIT PROTOCOL

FO K M O B I L E D I S T R I B Ư T E D DATABASE S Y S T E M S

Le H u u L a p

pnsts T e le co m m u n ica tio n s 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 t i o n

Recent e x p lo s io 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 v o lo p m e n t on m o b ile tra n s a c ỉio n s e s p e c ia lly on d is tr ib u tc d d a tab ase 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 to p o f w ire d -c u m -w ire le s s

n rtv v o rk w h i(‘h c o n s is ts o f m o b ile devices (such as PDA, ia p to p , c*tc) ca lle d M o b ilc

l ĩ n i t 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

eh a n n e ls fo r M U s, w h ic h move and h a n d o ff fro m ce ll to c e ll A d is tr ib u te d tra n s íic tio n g e n cT a lly in vo lve s p a rtic ip a tio n <>f se ve ra l s ite s , 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 th e m m ay be in Nvirtíless n e tw o rk For the pu rpose o f tra n s a c tio n s tu d y , th e netvvork can be vievved as a lo g ica l mess ( K ig u re 2)

Such M I)D B S m u s t execute c o m m it p ro to co ls to e n s u re a to m ic ity a fte r the traiìsaction executions on data Typically the commit protocols are based on

m rssage p a ssin # betvveen p n rtic ip n n ts in th e tra n s a c tio n , in one o r m ore d iffe re n t

phases such as 1PC (O ne-P hase O o m m it), 2 P C (T w o-P ha se C o m m it) and 3PC

(T hre e-P ha se C o m m it) |4 5, 10) U n lik e w ith fíxe d n e tw o rk f tho se p ro to co ls m ay nnt w o rk w e ll in vvireless n ie d iu m and w ith th e r ie x ib ilit y o f User m o h ility fo r

v a rio u s reasons |8, 10j:

• w ire le s s c o n n e c tio n is pro n e to ía ilu r e due to th e n a tu r c o f ra d io

p ro p a g a tỉo n lik e fa d in g , ob stacle and in te ríe re n c e ,

• M U has lim ite d b a tte r y po w e r, 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 c h 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 la y is h ig h e r due to u n re lia b le p h ysica l

co n n e ctio n s, and

• fo r d a ta b a se P a r tic ip a tin g site s are n o t u s u a llv in th e sam e n e tw o rk

p a r titin n , th e y are c o n n e c tin g fo r a sh o rt đ u ra tio n

V N U JO U R N A L OF S C IE N C E N at ■ Sci & Tech ĩ XIX, N.j4 2QQ3 _

3 5

Le H u u L a p, N gu yên K h a c I J c h 9 T r a n The T ruyen

Ilì th is p a p e r we e v a lu a te a new co m m it p rotoco l • C P M (C o m m it P rotocol fo r

M o b ile ) th a t is d o si^n e d to o ve r come some lim ita tio n (>f th e m ost vvrdelv used 2PC and to a llo w u s e r-e n a h le d d is c o n n e c tio n s and o fflin e Processing S e ctio n 2 p ro vid e s

a s h o rt re v ie w o f s e v r ra l c o m m it p rotoco ls and p o in ts out d ra w b a c k s o f 2PC in

m o b ile c o m p u tin g S ection 3 de scribes and a n alyzes 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 sectio n s p re s e n t th e m o d e lin g and s im u la tio n o f th e C P M

2 C o m m o p itm e n t in d i s t r i b u t e d tr a n s a c tio n Processing

2.1 T r a n s a c t io n M o d e ls

T h e re ha ve been m a n y 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] t h a t c a p tu re c la s s ic a l tra n s a c tio n m a n a g e m e n t íe a tu re s such as c o n cu rre n cy

c o n tro l a n d re c o v e ry |1Ị and m o b ility Fo r th e pu rpose o f t h is p a p e r, 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 t e d a t a s ite co n 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 tr ib u te d to p a r tic ip a tin g s ite s to execute

Th e g e n e ric tra n s a c tio n m odel con sidered in th is p a p e r c o n s is ts o f a set o f

a p p lic a tio n s (A P P k/ fo r a r b it r a r v k e J} o r ig in a tin g tr a n s a c ti o 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 tr o lle r T C J ha ndle s

c o m m a n d s re la te d to tra n s a c tio n on b e h a lí o f p a r tic 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 c a tio n o f T O in th e netvvork an d w it li

re spect 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 s u re s a tis fa c to rv p e rfo rm a n ce

2.2* C om m it p ro to c o ls

Kigure 1: Generic physical model o f MDDBS Figure 2: Logical view o f DDBS

It is a s tr ic t re q u ire m e n t o f a to m ic ity in da ta b a se tra n s a c tio n s , th a t a tra n s a c tio n T m u st e ith e r f u 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 s u re th is p ro p e rty a c o m m it p ro to c o l is exeeuted hy th e 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 f o r

; if t e r a ll > itfs in lu r m 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 l V ) p ro tn m l w h ií h \ve w ill d e s c rib r in m o re cit»f;ii 1 in thí* next

s u h srcĩion Kvrn thuimlì 2IH* IS r o n s i d e r c d to be i n s u í ĩ i c i r n t b c r a u s e C)f high

c o m m it ro s t «1 1 1 ( 1 p o ír u t lỉil (>f 1/0 h ỉo ck 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 r n 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 an d C o m m u n ica tio n s, in w h ir h P rC (P re su m e d

C o m m it) and K a r lv P rcp a rc |w 1 ]| an* o f thoso n o ticeab le O th e r re se a rch in c lu đ e s

1 IH 1 and v a r ia n ts such iis ( 'L P (C o o rđ in a to r Log P roto col), and ỈY V P ( Im p lic it Yes-

Voti* P rotocol) 1 PC Itf!iurt\s th e P R E P A R K phase o f ‘2PC b u t its a s s u m p tio n s a re too

s tr ic t to be tis H u l in ro m m e r ria l svstem s ỉn c o n tra s t to those a p p ro a ch e s, 3P C

(th re e -p h a s e 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 t o 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 th e 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 th e log, w r itc s th e lug to s ta b le sto ra g e and s ta r ts 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 vvhe the r o r n o t to c o m m it

l f th e 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 ' 1 adds re cord « R E A D Y T „ » to th e lo g a n d forces th e log to

th e s ta 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 po n re c e iv in g R E A D Y T,, fro m a ll TC ' w it h 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 simhIs 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 ' re ce ive s a

C O M M IT o r A B O R T fro m c o , it vvrites the message to th e lơg, th e n a ckn o w le d g e s back to CO

D esp ite th e p o te n tia l o f b lo c k in g , th e 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 u st‘d in p ra c tire fo r (lis trib u te đ da ta b a sc 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 tr ib u te d e n v iro n m e n t F o r e x a m p le , th e

m obile users m ay te m p o r n rily clisconncct to process its tra n s a c tio n fra tfm e n t o fflin e

fo r s a v in g povver, and th is a ctio 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 q u e n cy T h is is because o f a c r itic a l p o in t in 2PC, th a t a fte r tra n s a c tio n

e xe cu tio n s fin is h , t h i‘ (lecisio n to c o m m it re lie s e n tir e ly on th e chosen c o o rd in a to r,

w h ic h m ay no t hv iiNvarc oí’ m o bile User p re ỉere nce Once th e (.’ () decides to c o m m it

th e co n n e ctio n s need to biĩ kept to a vo id 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 th e vvireless

Le H u u L a p y N g u yê n K h a c L i c h y T r a n The T ru y e 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

th e y have to m a in ta in a lo ca i log to h a n d le CO*s d e m a n d s a t a n y tim e A n o th e r

w eakness 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 a re r a th e r e xp e n s iv e in m o b ile

Communications in te rm (>f tim e (and po ssibly 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 r o to c o l for m obile tr a n s a c tio n m o d e ls

3.1 T h e C P M C o m m ỉt P r o to c o l

In th e lig h t o f above d iscu ssio n , it is re a so n a b le to m in im iz e th e need to

co m m u n ic a te over w ire le s s lin k an d th e 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 co m m it pro to co l c a lle d C P M (C o m m it p ro to co l

fo r A /o b ile T ra n s a c tio n s ) th a t avo id s th e possible d is c o n n e c tio n s d u r in g th e co m m it process In a d d itio n C P M e n able s u s e r-p ro a c tiv e đ is c o n n e c tio n a n d o fflin e Processing fo r a r b it r a r y a m o u n t u f tim e The d is c o n n e c tio n m a y be in itia t e d b y

M U ’s tra n s a c tio n c o n tr o lle r b e fore decision to c o m m it is made

O u r c o m m it p ro to co l C P M o p e ra te s u n d e r g e n e ric 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 th e 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 v o 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 d iu m In C P M , 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 p la ys th e ro le o f c o o rd in a to r in 2PC d u r in g tra n s a c tio n e x e c u tio n Ít

d iv id e s th e tra n s a c tio n T, in to ừ a g m e n t set e x tra c ts it s íra g m e n t T lk and force-

\v rite s th e tra n s a c tio n log to s ta b le storage In s te a d o f s e n d in g th e re s t 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

c o rre s p o n d in g p a r tic ip a n ts an d s ta r ts a T R A N S _ E X E C tim e o u t Each p a r tic ip a tin g

s ite executes its p a rt a n d send c o 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 s e n d in g

C O M M IT _ R E Q U E S T message an d tra n s a c tio n log to c o U p o n re c e iv in g c o m m it

re q u e 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 sto ra g e and sends C O M M IT m essage to a ll p a r tic ip a tin g site s 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 an d acks back to c o ònce fin is h e d I f a ll such

a ckn o w le d g e m e n ts a re re ce ive d a t c o , th e c o 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 a b o u t 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 fo re h a n d T h u s

th e n u m b e r o f messages p a ssin g vvireless m e d iu m is m in im iz e d

O th e rw is e th e T R A N S _ E X E C tim e o u t is trig g e re d b e 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 an d 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 t o c o l fo r

m>i‘ds to c a r ry a ll n o c rs s a rv in fo r m ;itio n a b o u t th e a b o rte d tra n s a e tio n (i.e ID o f

p a r t in p iin t s 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 l l s its c o rre s p o n d in g p tơ a b o rt.

As m e iìtio n e d abọvo, th o p a r tic ip a n t P' and th e a g e n t T ơ a c tin g on its b e h a lf

a r r not n e c e s s a rily at t h i’ sanu' s i t t \ so d u r in g th e 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 an d, 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 th e existo nce o f “ c o m m it'’ log a t P \ I f th e lo g cỉoes n o t e x is t, the

T C ’ tu ll p to rtj d() th i' tr a n s a r tio n e x e cu tio n 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 co m p le te d T h e C P M can be expressed s e m i-fo rm a lly in

te rm o f pseudo code as in P ig u re 3 -F ig u re 6

A c ỉ i v i l i c s uỉ A P P :

b e g in

k

s e n d t r a n s a c t i o n T t- 0 TC

r-w a i t u r v t i l r e c é i v i n g i n s t r u c t i o n r r or ĩ ì TC

i f ( t h e i n s t r u c t i o n i s TRANS OK )

s t a r t n e w t r a n s a c t i o n a f t e r a t i m e i n t e r v a l

e ls e / / t h e t r a n s a c t io n 25 a b o r te d

r e s t a r t t r a n s a c t i o n a f t e r a t i m e i n t e r v a l

end

F ig u re 3: CPM a lg o rith m at a p p lỉc a tio n A PP

/\( 7/ v i l i c s cit T C :

B e g in

i f ( T C r e c e i v e a t r a n s a c t i o n T ; ) / / t h i s p l a y s t h e r o l e o f TC

b e g in k

TC s e n d T t o t:he s c h e d u l e r a n d t h e n g e t í r a g m e n t s { T t l |

T C " w r i t e a 11 « L O G T, » t o RAM

T C k è x t r a c t i t s p o r t i o n T lk

Ỳ

T C ' s e n d t h e r e s t ị T - T l k } t o a l l c o r r e s p o n d i n g p }

TC p r o c e s s i t s p o r t i o n T lk

TC s t a r t t i m e r a n d w a i t f o r c o n f i r m a t i o n f r o m { p ' }

i f ( t i m e o u t &Ẳ TC n o t r e c e i v e a l l T RANS_ ACK o f T )

b e g in

T C k s e n d ABORT T t o c o

T C k s e n đ TRANS NCT_OK T , c o A P P k

end

e l s e TC r o - e i v e 3 1 1 TRANS ACK:

4 0 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

b e g i n

T C k s e n d C OMMI T T ; , a l l < <L O G T n »

a n d i t s a c k o f f r a g m e n t t o c o

T C k 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 CO)

T C k s e n d T RA NS _ 0 K t o A P P k

end and

/ / F r o m now o n TC p l a y s t h e r c l e o f TC

e l s e i f ( T C ' r e c e i v e C O M M I T T i f r o m CO)

b e g in

TC s e n d r e c o r d « C O M M I T T , » a n d c o m m a n d C O M M I T T , t o 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

p-i f ( T C r e c e i v e c o m m i t c o n f i r m a t i o n o f p )

T C 1 s e n d C OM M I T _ A CK T j j t o c o

e l s e / / t i m e o u t

b e g i n / / p ; f a i l s

TC c h e c k l o g a t p-

i f ( e x i s t t h e l o g < < C O M M I T T 43» )

s e n d C O M M I T _ A C K T, t o c o

e l s e

T C : a s k p 3 t o r e d o t h e t r a n s a c t i o n p o r t i o n T j j

ba sed o n l o g f r o m c o

end

e n d

e ls e / / T C j r e c e ì v e ABORT T ịj f r o m c o

send ABORT T n t o p j

end

F ig u re 4: C P M a lg o rith m a t tra n s a c tio n c o n tro lle rs (T C k& T C 3)

A c t i v i t i e s a t P :

b e g in

i f ( p r e c e iv e t r a n s a c t i o n f r a g m e n t T , f r o m TC )

b e g in

e x e c u t e T n

s e n d T R AN S_ A CK T n b a c k t o TC*

w a i t f o r c o m m i t f r o m c o r r e s p o n d m g T C :

end

e ls e i f ( p r-e c e iv e C OM M I T T fro m TC )

b e g in com m it th e p o r t i o n T n

f o r e - w r i t e l o g < < C O M M I T T M > > t o s t a b l e s t o r a g 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

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

K v a l u ơ t i o n o f C P M (t c o m m it p r o t o c o l f o 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 t r■■ :m TC*)

i f ( r < ĩ 'G T • * »?x i s t _ )

• • x t r i ■’ : i : ‘ 4' i p a n t I L f r * r m e s s a g e A b OR T T.

e ls e

e x a m i n e <<L.OG T ; > > t o g e t T C 1

s e n d ABORT Tj t o a l l T C ’

e l s e i f (CO r e c e iv e a l l < < L O G !![,>>,

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

i f ( C O r e c e i v e a i i C C M MI T _ A C K T f r o m TC )

c o m p i e t e t h e c r a n s a c t i o n T, a n d

s e n d CO M M I T ACK T, t o T C'

en d end

F ig u re 6: CPM a lg o rith m at c o o rd in a to r c o

3 2 C ost a n a ly s ỉs <)f C P \ Ỉ

The C PM is designed w ith the cost m in im iz a tio n in m in d , 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 e ssa g e s L o g fo r c c -

w r it e s

C o m m u n ic a t io n

d e la y

w h e re 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 an d d is c o n n e c tio n d u r in g

c o m m ittin g , b u t whi*n too m a n v 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 a vo id process o v e rflo w T h e

c u to ff m e ch a n ism can be as s im p le as ro u n d -ro b in to rem ove th e ơ ld e st 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 o d o lin g

In a d d itio n to q u a lita tiv e a n a ly s is pre se n te d in p rc v io u s sectio ns 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 P M in d iffe re n t scen ario s and in

4 2 Le H ư u L a pt N g u y en K h a c Lichy T r a n The T r u y e n

co m p a riso n w it h th e 2 P ( \ Wi* doveloped a D iscre te K ve n t 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 ide a b e h in d th e DKS a p p ro a ch to s im u la tio n is based on co n ce p t o f e ve n t,

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 r re n t t i 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

a ss o c i a t e d w i t h t h 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 r re n t m odel <>f m o b ile C om m unications T h e c o m m it p ro to co ls are

o p e ra tin g on th e a p p lic a tio n la v e r w h e re th e u n đ e rlv in g n e tw o rk p ro v id e s th e tra n g p a re n t c o m m u n ic a tio n c h a n n e 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 it r a r y

n e tw o rk to p o lo g y b u t fo r th e 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 g e n e ric node vvith p r im a r y 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 e chan ism fo r ch e ckin g fo r e r r o 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 a l in c o m m u n ic a tio n n e tw o rk , th e loss o f message and

d is c o n n e c tio n a n d re c o n n e c tio n in w ire le s s lin k are m odeled as s u ffic ie n tlv long

d e la y 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 co n g e stio n s occur S im ila r ly h a n d o ff is a lso m odeled as

lo n g đ e la y vvith th e a ve rag e d e la y tim e fo r h a n đ o ff is ls T a k in g in to account th e

fa ct th a t a ve ra g e v e lo c ity o f m o b ile 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 a b o u t one every lOOs

Each s ittí c o n sists o f an a p p lic a tio n APP , a tra n s a c tio n c o n tr o lle r T C and a

da tab ase (o r p a r tic 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 th e 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 r tic ip a n t in th e m *tw o rk ra n d o m ly , t*xccpt fo r th e

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 s tc d a t c o an d p a r tic ip a tin g s ite 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 p o ssib le Processing o ve rlo a d th e tra n a a c tio n q u e u e a t c o op e ra te s

in ro u n d -ro b in s tra te g y , w h e re th e 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 f u ll

A t each s ite , C P U e x c c u tio n a n d 1/0 ciisk access are im p le m e n te d in m u tu a l

e x c lu s iv e fa s h io n ta k in g a c e rta in a m o u n t <)f timc* d e p e n d in g on n a tu re o f

o p e ra tio n s O th e r re q u e sts have to w a it u n t il thc» c o m p le tio n o f th e p re vio u s

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 a ve rag e P rocessing tim e

We use th e m ean 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 show n 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 t o c o l f o r 43

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

T r;m s :ic tio n S r ^ iìir n t K xiTU tion 34ms

T ra n s a c tio n Díĩgrec o f (iis trih u tio n 7-10 íra g m e n ts

Averagt* w ireless tra n s m is s io n de lay lOms

Àvi»ratft‘ vviređ tra n s m is s ío n đelay 5ms

Database system degrue o f d is trib u tio n >= 10 sites

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 ilit y

C u r r e n tly it s u p p o rts 2PC and C PM w ith s im ila r vvo rkin g 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 e r im e n ts a n d R esu lts

W t' have c o n d u cte d a co m p re h e n 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 ca sure d u s in g m e tric oí* tra n s a c tio n th ro u g h p u t p e r I in it o f tim e O th e r

fa c to rs in clu d e p o rtio n o f successful tra n sa ctio 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 r n a r o u n d tim e (F ig u re 10) T h e 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 th e s ta r t a n d th e end o f th e 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 presen ted in th is p a p c r a re o n ly those o f th e m ost in te re s t

F o r the syste n 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 , b u t fo r pu rpo ses of p e río rm a n ce s tu d y u n d e r h e a vy loads, th e re is o n iy one CO is set fo r th e w h o le s im u la tio n process, fo r bo th cases o f 2PC a n d C P M Fo r most stuclies, th o netvvork con sists o f 10 DBS b u t th e s im u la to r yllovvs a r b it r a r y

n u m b e r o f DBS

5.1 P e r fo r m 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 it io 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 c tiv e M ư s T h e syste m 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 ire le s8 lin k o f 0.5% • in n o rm a l s itu a tio n The queue size is 10 s e 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 igu re 7) A í t r r th a t p o in t th e system th r o u g h p u t ()f C P M decreases

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

ra p iđ ly w h ile th e 2VC p e río rm a n c e degrades ra th e r g ra d u a lly F o llo w in tf th e p o in t

w h e re 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 o rse th a n 2PC does

F ig u re 9 show s th e a ve rag e nu 111 be r o f message ()f th e tw o protoco ls T h is re a ssu re s

p re v io 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 s e 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 is c o n n e c tio n ra te 4%) T h is v riìì be

s tu d ie d in thc» n e x t sub section

Number of active MUs

F ig ure 7: Períbrm ance o f 2PC and C P M

under load vvịth n o rm a l disconnection ra te

(0.5%)

F ig u re 8: P erío rm a nce o f 2PC and C P M

u n d c r load w ith lìi^ h disconnection ra te

(4%)

Number of active paraftel active MUs

F ig u re 9: A vera gc n u m b e r o f messages per

tra n s a c tio n

3500 Average transaction turnaround time

F ig u re 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 l i n h f a ilu r e s

The com m on p ro p e rty betvveen h a n d o ffs and w ire le s s lin k fa ilu re s are long

d e la y (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 (1- lOms) H o w e ve r, th e a ve rag e 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 , whi<rh is

m 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 h o 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 (o r 3 6 k m /h ) across the ce ll d ia m e te r o f 300m T h e re s u lts

c o n ĩirm o u r in t u it io n as th e re is no s ig n iíìc a n t d e g ra d e in o v e ra ll syste m

th ro u g h p u t as h a n d o ff ra te grow s