DATA INTEGRITY - Database Design and Development: An Essential Guide for IT Professionals - page 469

Execute read operation. Set Read-timestamp(D) to the larger of Read-
timestamp(D) and Timestamp(T).

(2) T issues write(D)
(a) Timestamp(T) < Read-timestamp(D)
T asks to write D that has already been read by a younger or later transaction.
This means that a later transaction is already using the value of D and it would
be incorrect to update D now. T is too late to attempt to write D.
Reject write operation. Abort T and restart it with a new timestamp.
(b) Timestamp(T) > Write-timestamp(D)
T asks to write D that has been updated by a younger or later transaction. This
means that T is attempting to override the value of D with an obsolete value.
Reject write operation. Abort T and restart it with a new timestamp.
(c) Otherwise, when (a) and (b) are not true
Execute write operation. Set Write-timestamp(D) to Timestamp(T).

To illustrate the basic timestamp-ordering protocol, let us consider two transac-
tions T1 and T2 concurrently accessing savings and checking account records.
T1 wants to read the two records, add the balances, and display the total. T2 intends
to transfer $1,000 from a savings account to a checking account. Assume that T2 is
submitted after T1 so that Timestamp(T1) < Timestamp(T2). Figure 15-25 presents
the functioning of the timestamp-ordering protocol.
The basic timestamp-ordering protocol ensures conflict serializability. It produces
results that are equivalent to a serial schedule where transactions are executed in
their chronological order. In other words, the executions of operations would be as
though all operations of one transaction are executed first, followed by all opera-
tions of another transaction that are executed next, and so on, with no interleaving
of operations. A modification of the basic timestamp-ordering protocol affords
greater potential for concurrency. We discuss this in the next subsection.

Thomas's Write Rule Consider concurrent transactions T1 and T2 accessing data
item D. Assume that T1 starts earlier than T2 so that Timestamp(T1) < Time-

Transaction T1

Execution under timestamp
ordering protocol

T1

READ (SAV)
READ (CHK)
SUM :

TIME T1 T2 CHK SAV SUM

= (CHK) + (SAV)

t1 READ (SAV) 2000 15000

DISPLAY SUM

t2 READ (SAV) 2000 15000

t3 (SAV) : = (SAV) - 1000 2000 15000

t4 WRITE (SAV) 2000 14000

Transaction T2 t5 READ (CHK) 2000 14000

t6 READ (CHK) 2000 14000

T2
READ (SAV)
(SAV) :

t7 SUM : = (CHK) + (SAV) 17000

t8 DISPLAY SUM

= (SAV) - 1000

t9 (CHK) : = (CHK) + 1000 2000 14000

WRITE (SAV)
READ (CHK)
(CHK) : = (CHK) + 1000
WRITE (CHK)
SUM : = (CHK) + (SAV)
DISPLAY SUM

t10 WRITE (CHK) 3000 14000

t11 SUM : = (CHK) + (SAV) 17000

t12 DISPLAY SUM

Figure 15-25 Timestamp-ordering protocol.