You are on page 1of 8

A model for transaction

management in mobile databases


Ziyad T. Abdul-Mehdi, Ali Bin Mamat,
Hamidah Ibrahim, and Mustafa Mat Deris

© IMAGESTATE

he evolution of mobile computing devices and wireless base. We present a model that implements this framework in

T networks has created a new mobile computing environ-


ment. Users equipped with a mobile host can access,
retrieve, and process information while in mobility.
Laptops become more powerful data processing elements.
Traditional transaction model has moved forwarding to mobile
an asynchronous and synchronous system.
In recent years, several research articles regarding dis-
tributed databases have been published (Abdul-Mehdi Z.T
et al., 2007; P. Padmanabhan et al., 2006; S. Bottcher et al.,
2006; M. Mat Deris et al., 2004; D. Agrawal and A. El-Abbadi,
transaction. This paper proposed the model with the aims at 1996; J. Holliday et al., 2002; and P.A. Berstein et al., 1987).
solving the stated issues. A key requirement in such an envi- These articles revealed that data replication management
ronment is to support frequent disconnection of mobile data- is one of the current issues in distributed database that
has yet to be solved. It was on this basis that this study
Digital Object Identifier 10.1109/MPOT.2010.936929 was initiated.

32 0278-6648/10/$26.00 © 2010 IEEE IEEE POTENTIALS


Table 1. Weakness of transaction models.

No. Transaction Managements Models Problems of the Models

Two-tier replication • Executes transaction without the knowledge of what other transaction does.
1 (J. Gray et al., 1996) • The commit time of a transaction at MH is large.
Clustering • Weak transaction can proceed when the network is partitioned and this will cause
2 (E. Pitoura and B. Bhargava, inconsistent data, so the transaction will abort.
1998)

Kangaroo • Break the consistency and durability of the transaction.


3 (M.H. Dunham et al., 1999) • Break atomicity in split mode.
• The conflict resolution mechanism is only so robust and the system will default to
IOT
4 manual repair in some situations. This case will take a large time commitment or the
(Rich Kong et al., 2001) transaction will abort because it does not satisfy.

Planned disconnection modes


5 • The model only allows one MH to have a copy of the data object.
(J. Holiday et al., 2002)

Several works have been completed previously disconnected. Each MH is each data item Di is calculated using
in the area of mobile databases as allocated di a portion of di and the bal- function fi 1 di,ni,mi 2 , which will be
shown in Table 1. This continuing effort ance of di, say bi, is kept at the BS. The described in detail later.
has one common goal to achieve: to value di is an integer and 0 , di , di. The purpose of di is to determine
find a suitable transaction management Obviously, d i 5 bi 1 1 n i 1 m i 2 * d i . We whether a request (transaction) to
model for mobile databases. The study can view d as a total maximum changed update or change a data item Di is
on the previous models discovers some allowed at each MH and at the same accepted or not. If the transaction
weaknesses or issues, as shown in time as a partial replication of data item. changes the data value by at most di
Table 1, which need to be solved. The Thus, di is the total maximum changed in a MH, then it is free to precommit.
issues include: of Di allowed at MH. The value di for It does not have to wait for results of
• only one mobile host (MH) is al- the base transaction at the BS. This
lowed to update the data item. reduces the transaction’s commit time.
• large number of rejected transactions If the change of data value is more
• commit time of transactions at MH than di , the transaction is not executed
is large.
An MH can disconnect from at MH, instead it is sent direct to the
the BS during the timeout server. This transaction is called a
The system model period and can perform request transaction.
To explain our model, a simple updates. To control the validation of di, the
architecture, as illustrated in Fig. 1, is BS assigns timeout parameter t, whose
sufficient. It consists of one base station value indicates a duration within which
(BS) and a number of MHs, which com-
municate with the BS through a wire-
less network.
The main idea is that transaction LAN
execution can be done at the BS and
MHs. Transactions at an MH can update
data locally and then precommit. When Network Line
the MH connects to the BS, these pre-
committed transactions are sent to the
BS and re-executed as base transactions
(BT). BTs are serialized on the master
copy of the data stored at the BS. This
will result in data consistency.
The BS holds and manages master
data. Data is distributed to MHs and this Base Station (BS)
data can be updated by MHs. Let D be
a set of data items such that
D 5 5 D 1, D 2, c, D n 6 , where i [ N, set
of natural numbers (nonnegative inte-
g e r ) . F o r s i m p l i c i t y , D i [ N. D i
represents a data item related to a dis- Mobile Host
tributed reservation system such as (MH 3)
number of flight seats available, or num- Mobile Host
ber of football tickets. Let di be the (MH 1) Mobile Host
(MH 2)
current value of Di and let n i be the
number of MHs requesting a data item
Di. Let mi be the number of MHs Fig. 1 A simple architecture for mobile databases.

MAY/JUNE 2010 33
the value of di is valid. The timeout Taking the current example, d i 5 75, An MH has 1 d i, d i 2 for every Di it
value for di should be appropriate so n i 5 2 and this will give us has cached and their t. MH executes
that an MH can send their updates with- transaction ti . Let di2ti be the change
in duration of t. The BS will not update X server 5 300 2 1 75 * 2 2 made by ti to Di . Let C be the current
the value of the data until t has elapsed. Xserver 5 150. value of the total change in Di (since
We assume that the MH reconnects and the last transfer of value di ), C is the
send their updates before the timeout Transaction execution accumulation of successful commit
expires at the BS. at base station transaction at MH. Hence, more than
An MH can disconnect from the BS The BS will perform the main tasks one transaction can precommit at an
during the timeout period and can per- including: MH while it is in a disconnection
form updates. If the MH disconnects for • Calculating di for each data object. mode, the value of C is initially 0 for
a period longer than the timeout, when • Associating to each di a timeout every disconnection. If transaction ti
it reconnects it will send the precommit value t. changes the value of Di by any
and request transactions both as request • Executing transactions on a mas- amount di2ti, then the amount di2ti is
transactions. ter data item in order to maintain data added to C .
consistency. The possible cases are as follows:
Data allocation Once di is calculated, BS transfer If di2ti ,5 d i and C ,5 d i then the
A data value di is a dependent value 1 d i, t 2 for each data item 1 Di 2 to each transaction ti is committed at MH and it
and distributed among the BS and MHs MH. During a timeout t, the BS can is sent to the BS for re-execution on the
using a specified function. Let di be a receive either precommitted transactions master data.
data value allocated to an MH and ni (transactions that have made an amount If di2ti ,5 d i and C . di then the
be the number of the new MHs that of updates within the limit di on the MH transaction ti is blocked at MH
require the data value. Let mi be the and committed) or it can receive request and submitted to the BS as a request
number of MHs that have been allocat- transactions (transactions directly sent to transaction.
ed the data value and disconnected. the BS by the MH). A request transac- If di2ti . dx then the transaction ti is
The value of di is calculated by using tion (a transaction violating the limit or blocked at MH and submitted to the BS
the following function: timeout t is sent to the BS as a request as a request transaction.
transaction for execution on the master
di 5 fi 1 di,mi,ni 2 database) is not executed at an MH. Proposed model examples
5 : 1 AV 1 1 e * r 2 2 * d i/ 1 m i 1 n i 2 ; (1) The transaction would have changed First consider a data item X repre-
the value of di by more than 1 di 2 at the senting the total number of football
where di is the data value amount allo- MH. The execution of transactions, pre- tickets (X belongs to natural number).
cated to every MH wish to disconnect commit and request transaction on mas- Assume that the initial value di of X is
and AV is an average, and the average ter data, is done as follows: 300 and ni the number of MHs that
usually is 0.5, the mean idea for the AV • The BS serializes the precommit- want to disconnect simultaneously, is 2.
is discussed in the methodology of this ted transaction according to the order of There is no previously data allocation to
research. arrival at the BS, the value at the BS will any MH (i.e., mi 5 0 ). Using the data
The variable r represents the be updated by the expression: allocation scheme defined previously,
1 r 1 1 2 th request of the data item value we obtain:
by MH, initially r is zero and increased X server 5 X server 1 1 d i 2 C 2 $ 0 (3)
by one whenever there is a new
di 5 311 1/2 2 1 1 0.05 * r 2 * di 2 /
request. However, the increment of r where C is current value of total change
1 mi 1 ni 24
will stop after reaching the maximum at MH.
value of 9. • Once precommitted are complete- 5 1 1 1 1/2 2 1 0 2 * 300 2 / 1 0 1 2 2
For instance, we have two MHs ly executed, request transactions are 5 75
request the data value of an item whose then executed in their order of arrival. X server 5 X server 2 1 d i * n i 2
initial value is 300, and assume that this The BS send message to the MH wheth- 5 300 2 1 75 * 2 2
is the first group of MHs request for the er the transaction was committed or 5 150.
data. Thus r 5 0, m i 5 0, n i 5 2 and d i aborted. The value of the X server will be
is given by updated according to the expression: This means that MH1 and MH2 will
have the privilege to update data item
d i 5 3 1 0.5 1 0.05r 2 * d i/ 1 m i 1 n i 2 4 X server 5 X server 2 1 d i2ti 2 d i 2 $ 0 (4) Di (not more than the maximum of 75).
Fig. 2 illustrates the allocation of data
d i 5 3 1 0.5 1 0.05 * 0 2 * 300/ 1 0 1 2 2 4 where di2ti is current value of request item Di to MH1 and MH2 and the bal-
di 5 75. transaction at MH. ance, X server which is used for another
MH wanting to disconnect during MH1
Let X Server be the current data value Transaction execution and MH2 .
maintained at the BS. Initially at mobile host Throughout, MH1 and MH2 discon-
X Server 5 300. After allocating each MH Essentially, an MH can accept the nection, another MH3 would also want
withd i , we obtain the new value of change and precommit if the change is to disconnect. By using the data allo-
X Server given by within the limit of change, otherwise cation function and increase r by 1
the transaction is sent to the BS as a after first disconnection, BS will
Xserver 5 Xserver 2 1 di * ni 2 . (2) request transaction. allocate to MH3 some amount (but not

34 IEEE POTENTIALS
BS BS

Xserver = 150 Xserver = 123


3 Wish to
∑δi = 177 Disconnect
i=1
MH1 MH2 δi = 75 3

MH1 di =Xserver + ∑ δi
Already i=1
Disconnected
δi = 75 δi = 75 MH3
δi = 27
δi = 75
MH2
Fig. 2 MH1 and MH2 wish to disconnect
from BS.

Fig. 3 MH3 wish to disconnect after MH1 and MH2 disconnected.


the same amount that taken from MH1
and MH2 ).

di 5 311 1/2 2 1 1 0.0.05 * r 2 * di 2 /


1 mi 1 ni 24
5 111 1/2 2 1 1 0.05 * 1 22 * 150 2 / BS
1 2 1 1 2 527 Xserver = 123
X server 5 X server 2 1 d i * n i 2 δx (MH1,2) = 75
5 150 2 1 27 * 1 2 5 123. δx (MH3) = 27
3

So, the MH3 will disconnect with max- ∑δi = 177


i=1
imum amount 27, as shown in Fig. 3. 3

Now, consider transactions t1, t2, and di = Xserver + ∑ δi


i=1
t3 arrived to MH1, MH2 , and MH 3
respectively, and suppose t1 books 60
tickets, t2 books 100 tickets, and t3
books 35 tickets. Let Ci represent Mobile Disconnected
change in value of data object Di. Each
MH that has an allocation of the data
item Di will maintain the value of C.
The execution of these transactions δi = 75
δi = 27 δi = 75
at each MHs is as follows: δi–t2 = 100
MH1 : Initially C 5 0 , di 5 75, t1 books MH3 δi–t3 = 35 Ci = 60
6 0 t i c k e t s , s o di2t1 5 6 0 a n d MH1
MH2
Ci 5 Ci 1 1 di2t1 2 5 6 0 . As Ci , di, t1 is
committed at MH1 and is sent to the BS
for re-execution as shown in Fig. 4.
MH2 : Initially C 5 0. t2 books 100 Fig. 4 Transaction processing at MHs.
tickets, so di2t2 5 100 and Ci 5 Ci 1
1 di2t2 2 5 100 and Ci . di (As shown in
Fig 4). ance of di is added to the value at BS as ing t2 and t3 is given below and shown
MH3 : Initially Ci 5 0. t3 books 35 shown below: in Fig. 5.
t i c k e t s , s o di2t3 5 35 , Ci 5 Ci 1
1 di2t3 2 5 35, and Ci . di (As shown in X server 5 X server 1 1 d i 2 C i 2 X server 5 X server 2 1 d i2ti 2 d i 2
Fig 4). After all the assumptions above, Xserver 5 123 2 1 60 2 75 2 X server 5 138 2 1 100 2 75 2 5 113
consider that five cases may occur for Xserver 5 138. Xserver 5 113 2 1 35 2 27 2 5 105.
this scenario.
Transactions waiting for execution Case 2: MH3 reconnects before MH1
At base station are maintained in the queue in their a n d MH1 r e c o n n e c t s b e f o r e
Case 1: MH1, MH2, and MH3 reconnect order of arrival. For this case, t2 will MH2 1 t3 S t1 S t2 2 .
at the same time to the BS. Transaction t1 enter the queue before t3. Thus, t2 is In this case, Fig. 6 shows MH3 will
is precommitted and t2 and t3 are request- executed before t3 1 t2 S t3 2 . If the BS send the request value of di2t3 first to
ed transactions. The BS will receive the has enough value of data item then it process at the BS but if the BS is busy,
transaction t1, t2, and t3 in this order. t1 will process t2 and t3 and sends a com- then transaction t3 will enter the queue.
will be re-executed first before t2 and t3. mit message to MH3—if not, it will send If t1 arrived when t3 is being processed
Transaction t1 is committed and the bal- an abort message. The effect of execut- at the BS, then t1 will wait in the queue,

MAY/JUNE 2010 35
otherwise it will process directly. If MH2 X server 5 123 2 1 35 2 27 2 5 115 Case 3: MH1 completes the execution
reconnects, it sends di2t2 to the queue. of transaction t1 and before the timeout
X server 5 X server 1 1 d i 2 C i 2
All these transactions are executed in the t is expired, it receives t4, which books
order of t3, t1, t2 (t3 S t1 S t2). The effect X server 5 115 1 1 75 2 60 2 5 130 for 20 tickets. So, Ci . di. The current
of executing t3, t1, and t2 is given below: X server 5 X server 2 1 d i2t2 2 d i 2 value of C is 60. After adding the new
request, C is 80. Thus, the request
X server 5 X server 2 1 d i2t3 2 d i 2 X server 5 130 2 1 100 2 75 2 5 105 transaction t4 for MH1 will be submitted
to the BS when it reconnects. It will
send t4 as a request transaction to the
Xserver = Xserver – (δi–ti – δi) queue after sending t1 to re-execute at
Xserver = 138 – (100 – 75) = 113 (After Execution of t1) the BS. Fig. 7 illustrates the execution
of t1 and t4.
Xserver = 113 – (35 – 27) = 105
Head of Queue
X server 5 X server 1 1 d i 2 C i 2
Queue BS
X server 5 123 1 1 75 2 60 2 5 138
t3 t2 t1
X server 5 X server 2 d i2t4
Xserver = Xserver – (Ci – δi)
Xserver = 123 – (60 – 75) X server 5 138 2 20 5 108.
Xserver = 138 (Execution of t1)
Xserver = Xserver – (δi–ti – δi)
Xserver = 138 – (100 – 75) = 113 (Execution of t2)
Case 4: MH3 has expired its timeout t,
Xserver = 113 – (35 – 27) = 105 (Execution of t3) then the BS takes back the amount of
di and adds to X server. When MH3 recon-
nects, all the transactions will be treat-
First Transaction ed as request transactions at the BS.
t2 Will Be Executed Fig. 8(a) shows how the BS will take
t1 Second Transaction the data that distributed to MH3 back to
t3 Will Be Executed X server. Fig 8(b) shows how the BS con-
Third Transaction
δi = 75 siders all the transactions that come
after timeout t is expired from MH3 as a
Ci = 60
δi = 27 δi = 75 request transaction(s).
MH1
MH3 δi–t3 = 35 MH2 δi–t2 = 100 Case 5: MH3 reconnects and requests
more than the data item resources of di
at MH3 and X server at the BS. In this
case, the BS will abort the transaction
Fig. 5 Execution of t2 and t3. and the BS sends an abort message to
MH3 as shown in Fig. 9.

Proof of correctness
BS In this section, we give a proof of
Queue Head of Queue correctness of our model that sets limits
t2 t1 t3 on the change allowed in each data
Xserver = Xserver – (δi–t3 – δi) object value at MH that does not allow
Xserver = 123 – (35 – 27) = 115 (Execution of t3) transactions to violate this limit. During
Xserver = Xserver + (δi – Ci) the timeout 1 t 2 , the value of data object
Xserver = 115 + (75 – 60) = 130 (Execution of t1) Di in BS may diverge from the consis-
n
tent valueby a maximum amount Si51 di,
i

Xserver = Xserver – (δi–t2 – δi)


Xserver= 130 – (100 – 75) = 105 (Execution of t2) which the total value of data object Di
allocated to MHs during disconnection.
In this way, transactions are allowed to
First Reconnection t execute an inconsistent data object at
3 the BS but the inconsistency in data
n
value is bounded by Si51 di. Our model
i

t1 Second Reconnection includes the calculation of di for each


t2 Third Reconnection data object Di and also lists steps for
processing transactions of MHs.
If the MHs violates the time out of di,
δx = 27 δx = 75 δx = 75
the transaction will be blocked and
MH3 δi–t3 = 35 δi–t2 = 100 MH1 Ci = 60
MH2 when the MHs reconnect to the BS, the
MHs will send the blocked transaction
as a request transaction to the queue of
BS. Transactions enter the queue in
Fig. 6 Execution of transaction in order of t3, t1, t2. their order of arrival.

36 IEEE POTENTIALS
For the subsequent discussion, the
definitions of the history and serializa- BS
tion graph (SG) are needed. A history is Head of Queue
Queue
a partial order of operations that repre-
t4 t1
sents the execution of a set of transac-
tions. Any two conflicting operations Xserver = Xserver + (δi – Ci)
must be comparable. Let H be a history. Xserver = 123 + (75 – 60) = 138 (Execution of t1)
The serialization graph for H, denoted Xserver = Xserver – δi–t4
by SG(H), is a directed graph whose Already Xserver = 138 – 20 = 108 (Execution of t4)
nodes are requested transactions in H Disconnected
and whose edges are all ti S tj 1 i 2 j 2 , Transaction t1, t4
such that one of ti’s operations can pre- δi = 27 t1, t4 Reconnections
cede and conflicts with one of Ti ’s δi–t3 = 35
operations in H. To prove that a history
δi = 75
H is serializable, we only have to prove MH3 δi = 75
δi–t2 = 100 δi–t4 = 20
that SG(H) is acyclic. MH1
MH2 Ci = 60
We denote the arrival time of trans-
action ti by Tti , let t1 and t2 be two
request transactions of MH in a history
H of BS produced by the proposed
Fig. 7 Precommit and request transaction processing at the BS.
model. Assume that t1 and t2 comes
from the same MH and assume if two
MHs reconnect at the same time. The
first MH-sent transaction will be served BS
first. If there is an edge ti S tj in SG(H), Xserver = Xserver + δi
then Tt1 a Tt2 . Since ti S tj , the two Xserver = 123 + 27
transactions contain conflict operations.
Let w denote the update operation and Xserver = 150
r for the read operation. There are three
possible cases. Already
Disconnected
Case 1: w1 3 x 4 S w2 3 x 4
Suppose Tt2 a Tt1 . Therefore, t2 δx = 27
enters the queue before t1. If w1 3 x 4 is δi–t3 = 35
sent to the base station first, t2’s write Ci = 0
lock on x must be released before
w1 3 x 4 is sent to the base station to
MH3
update. If w2 3 x 4 is sent to the base sta-
tion first, it will either be processed
before w1 3 x 4 is sent to the base station, (a)
or be discarded when the base station BS
receives w1 3 x 4 , because w2 3 x 4 has a
smaller arrival time. Therefore, w1 3 x 4 is δi–ti = δi–t3 + Ci
never processed beforew2 3 x 4 in queue δi–t3 = 35 + 0
at the BS. Such conflict is impossible. δi–t3 = 35
This is a contradiction. Connect After
Time Expired
Case 2: r1 3 x 4 S w2 3 x 4
Send δi–t3 to BS
If t2 holds write lock on x when t1
δx = 27
requests the read lock (MH that wishes
δi–t3 = 35
to disconnect, so the BS will read
Ci = 0
X server , before allocating the value of
data item di to this MHi), therefore it
must be that Tt1 a Tt2 and t1 must com- MH3
(b)
mit before t2.

Case 3: w1 3 x 4 S r2 3 x 4 Fig. 8 (a) BS takes back the value of data item from MH3. (b) Execution of request
transaction after time out is expired.
Since t1 is already in the write phase
before t2 reads X server , we must have
Tt1 a Tt2 . Suppose there is a cycle Therefore, no cycle can exist in SG(H), tion of request and pre-committed
t1 S t2 S ....... S tn S t1 (H). By the so our model only produces transactions at BS and transferring of
above argument, we have Tt1 a Tt2 , serializable histories. Thus, the consis- the consistent value of the data object
. . . , Ttn a Tt1. This is impossible. tency restoration includes the execu- to the MH at the time or reconnection.

MAY/JUNE 2010 37
In the proposed model, both trans- Joey transaction (JT) causes the entire
BS actions (precommit and request) allow KT to be undone by performing com-
Xserver = 3 the data to be updated immediately. pensating transactions. This mode also
For the precommit transaction (i.e., the breaks Atomicity, Durability and
Reconnected local transaction in MH), if the request Consistency; and 2) Split mode: abort-
Send an for the data value is less than the ing the JT requires the termination of
Abort Message amount of the data value allocated d, the KT but will not undo the previously
δi = 27 the MH can make the update locally committed JTs.
δi–t3 = 35 and send the precommit transaction to Upon the decision of the local data-
Ci = 0 the BS to re-execute as the base trans- base management systems, the currently
MH3
action BT. This will decrease the exe- running subtransactions are committed
cution commit time for the transaction or aborted. This mode also breaks ato-
and the transaction will not reject. micity and the consistency. Therefore,
Fig. 9 Aborting transactions.
Re-executing the transaction at BS as none of the two mentioned modes
BT will keep the database at a consis- ensure serializability.
tent level at the end of the timeout for The aim of introducing the pro-
Compare and contrast any transaction, as the aim of this posed model is to achieve the consis-
The first model is known as the study is to keep the database value tency level, so as to ensure that only
two-tier replication model. In this consistent. Furthermore, when the MH the BT can change the master copy
model, the MH executes a transaction is delayed in reconnecting to the BS, when the transaction at BS as BT is re-
without knowing what other transac- the BS will return the data value executed. When the MH violates the
tions are doing, leading to a large num- amount allocated for this MH to the timeout t, then the BS will backup the
ber of rejected transactions. The com- X server in order to keep the data at the amount of data value d that is allocat-
mit time of a transaction at MH is large consistent level. ed to the MH in order to keep the data
because the outcome will only be at the consistency level. Although this
known after the BT has been executed procedure will interrupt one of the
and the result is reported back to the objectives of the study, this research
MH. In the proposed model, the data The problem with the aims to reduce the execution time and
to MH is not fully replicated but some isolation only transaction the number of rejected transactions by
amount value d for every MH to dis- model is that the conflict not depending on the consistency
connect and take permission in updat- level because if the data item does not
ing d locally are allocated to increase
resolution mechanism is consist defiantly, the transaction
the data availability, increase MH only so robust and the should be aborted.
autonomy, and reduce the execution system will default to The problem with the isolation only
commit time. The main task for the BS manual repair in some transaction (IOT) model is that the
is to calculate d, associate to each d conflict resolution mechanism is only
time out t and execute a transaction on
situations. so robust and the system will default
the master data item in order to main- to manual repair in some situations. In
tain data consistency in the proposed addition, this will take time for any
model. In this way, the BS is aware When the MH is reconnected to the transaction to commit another problem
and knows about every transaction BS, it will consider all the transaction(s) when the second class transaction
executed. In addition, the allocation of as request transactions even if the MH conflicts with the other second class
data value amount d helps every MH makes precommit (locally update) for transaction in the BS, leading it to re-
to make a local update. The model will these transactions. In this case, the ject the transaction. Otherwise, the
not reject any transaction if the limited reduction of execution time is violated model will adopt a specific application
d for MH or the amount of data item but it saves the data at the consistent to recover this problem, which is usu-
valve at the BS (if it is a request trans- level. Suppose the BS is allocated d to ally not resolved (it will reject the
action) if not violated. In this model, MH and MH is not reconnected to the transaction). The use of a queue that
the BS will keep the database item at BS after timeout expired, the MH will serializes the transactions as used in
the consistent level by re-executing the not reconnect anymore. The database the proposed model is suggested. The
transaction as BT to the master data- item at BS cannot be consistent if some new proposed model allows the up-
base item. of the amount is left in MH. In this case, date of data item locally, if the limita-
The clustering model defines two the transaction is not rejected but the tion of the data value amount that is
types of transactions: the weak and execution will be delayed. The BS is allocated to any MH is not violated.
strict transactions. The weak transac- associated with every allocation d time- The time to check the conflict be-
tion can proceed when the network is out to keep the database updated and tween the transactions such as in the
partitioned, but this will cause an the database item consistent. The MH IOT model will not occur in the
inconsistency level because the data- should respect the timeout associated proposed model as it avoids any trans-
bases state should be processed only for every allocation d. action conflicts from occurring.
by the strong transaction. Hence, the The third model, the Kangaroo Finally, the fifth model involves a
weak transaction will not allow chang- transaction (KT), reveals the weakness planned disconnection mode. However,
ing of the database state and this will of the two modes processing a KT; i.e., the problem in this model is highlight-
be aborted or will not be consistent. 1) Compensating mode: aborting the ed when every MH is disconnected, as

38 IEEE POTENTIALS
all the data updated may get lost cial applications such that time reduc- ad-hoc network databases,” VLDB J., vol.
because no BS will update the MH es along with an increase in MH reli- 17, no. 5, pp. 1143–1164, 2006.
when the MH reconnects. Reconnection ability and autonomy. • R. Kong, S. Lau, and D. New,
will cause the execution time of the “Mobile data access: Disconnected and
transaction to increase or the transac- partially connected approaches,” Dept.
tion after period of waiting time will of Computer Science, Univ. of Califor-
abort. Moreover, the system only The consistency and the nia, San Diego, La Jolla, CA, 2001.
allows one MH to have a copy of the • S. Bottcher, L. Gruenwald, and
data item and disallows the data to be
isolation management can be S. Obermeier, “An atomic web-service
available at multiple MHs. In the pro- a solution for many database transaction protocol for mobile environ-
posed model, the number of MHs commercial applications such ments,” presented at the EDBT-Work-
using our function f 1 di, mi, ni 2 are that time reduces along with shop Privacy Information Management
allowed and the function will allo - (PIM2006), Munich, Germany, 2006.
cate d for every MH to disconnect.
an increase in MH reliability
Therefore, the data item will not allo- and autonomy. About the authors
cate to only one MH but to the number Ziyad T. Abdul-Mehdi (Ziyad.tariq@
of MHs that wish to disconnect. mmu.edu.my) is a senior assistant profes-
In addition, the architecture of the sor at the Department of Information
new model allows the BS to allocate the Read more about it Technology, Sohar Applied Sciences
database item by allocating d with time- • Z. T. Abdul-Mehdi, A. Mamat, College, Sohar, Sultanate of Oman. He
out t, this also means that the BS is H. Ibrahim, and M. M. Dirs, “Replication holds a postdoctoral degree in security
aware and knows every change that management in mobile disconnection databases and a Ph.D. degree in comput-
occurs to the data item. Meanwhile, the mode,” in Encyclopedia of Mobile Com- er science (mobile databases) from Uni-
BS manages the master copy of this puting and Commerce (EMCC), Hershey, versity Putra Malaysia (UPM), Serdang,
data item by re-executing the transac- PA: IGI Global, 2007. Malaysia. He has published more than
tion as BT at the master data so as to • D. Agrawal and A. EL Abbadi, 20 research papers in journals and
maintain the data consistency. So, both “Using reconfiguration for efficient man- proceedings. His research interests in-
the rejected transactions and the com- agement of replicated data,” IEEE Trans. clude data transaction management and
mit execution time will be reduced Knowledge Data Eng., vol. 8, no. 5. pp data security.
because the proposed model does not 786–801, Oct. 1996. Ali Bin Mamat (ali@fsktm.upm.edu.
use the epidemic message to update the • E. Pitoura and B. Bhargava, “Data my) is an associate professor in comput-
data (an epidemic message takes time consistency in intermittently connected er science at UPM. He earned his Ph.D.
to commit the transaction). distributed systems,” IEEE Trans. Knowl- degree in computer science from the
edge Data Eng., vol. 11, no. 6, pp. 896– University of Bradford, U.K., in 1992.
Conclusion 915, June 1998. He has published more than 50 research
Transaction management models for • J. Gray, P. Helland, P. O’Neil, and papers in journals and proceedings. His
databases, in particular the centralized D. Shasha, “The dangers of replication research interests include databases,
databases, has long been established. and a solution,” presented at the ACM data grid, XML, and Web semantics.
Due to the nature of mobile computers, SIGMOD Conf., Montreal, Canada, 1996. Hamidah Ibrahim (hamidah@fsktm.
the traditional transaction management • J. Holliday, D. Agrawal, and A. El upm.edu.my) is an associate professor
models are no longer appropriate. In a Abbadi, “Disconnection modes for mo- at the Faculty of Computer Science and
mobile computing environment, the bile databases,” J. Wireless Netw., pp. Information Technology at UPM. She
atomicity, consistency, isolation, and 391–402, 2002. earned her Ph.D. in computer science
durability properties are not entirely • M. H. Dunham and V. Kumar, from the University of Wales Cardiff,
obeyed. Therefore, the need for new “Impact of mobility on transaction man- U.K., in 1998. Her research interests
transaction models for mobile databases agement,” in Proc. ACM Int. Workshop include databases (distributed, paral-
is inevitable. As a result, several models on Data Engineering for Wireless and lel, mobile, bio-medical, XML) focusing
have appeared, each with a strength Mobile Access, Seattle, WA, USA, ACM on issues related to integrity con-
and not without weaknesses. Press, 1999, pp. 14–21. straints checking, cache management,
We have proposed an alternative • M. Mat Deris, J. H. Abawajy, and integration, access control, transaction
model of transaction management H. M. Suzuri, “An efficient replicated processing, and query processing and
with the abilities to overcome some data access approach for large-scale optimization.
problems that have been identified. distributed systems,” in Proc. IEEE Int. Mustafa Mat Deris (mmustafa@kuittho.
One important aspect of our model is Symp. on Cluster Computing and the edu.my) earned his B.Sc. from UPM,
that it removes the restriction that lim- Grid, 2004, pp. 588–594. M.Sc. from the University of Bradford,
its the permission to update the data • P. A. Bernstein, V. Hadzilacos, and England, and Ph.D. from UPM. He is a
to only one mobile computer. The N. Goodman, Concurrency Control and professor of computer science in the
experiment result has shown that our Recovery in Database Systems. Reading, Faculty of Information Technology and
model functions correctly. MA: Addison-Wesley, 1987. Multimedia, UTHM, Malaysia. His re-
For this work, the consistency and • P. Padmanabhan, L. Gruenwald, search interests include distributed data-
the isolation management can be a A. Vallur, and M. Atiquzzaman, “A survey bases, data grid, database performance
solution for many database commer- of data replication techniques for mobile issues, and data mining.

MAY/JUNE 2010 39

You might also like