CIS 720

Concurrency Control

Locking

• Atomic statement– Can be used to perform two or more updates

atomicallyTh1: …. < x = x + 1; y = z>;…….Th2:………….<m = m + 1;….>;…….

Transactions

• A database system is a set of shared data objects

• A transaction is a sequential program which accesses data objects in the database

• Each transaction is a sequence of read and write operations

The Transaction Model

• Examples of primitives for transactions.

Primitive Description

BEGIN_TRANSACTION Make the start of a transaction

END_TRANSACTION Terminate the transaction and try to commit

ABORT_TRANSACTION Kill the transaction and restore the old values

READ Read data from a file, a table, or otherwise

WRITE Write data to a file, a table, or otherwise

Transactions

• Each transaction is a sequence of read and write operations

• The read set of transaction T, denoted by rs(t), is a set of variables read by T.

• The write set ws(T) is defined similarly

Banking SystemDeposit(amount, account)

{ x = db.account; x = x + amount; db.account = x; }

Withdraw(amount, account) { y = db.account;

if y > amount y = y - amount;

db.account = y; }

Distributed Transactions

BEGIN_TRANSACTION reserve MCI -> JFK; reserve JFK -> FRK;END_TRANSACTION

• A database has an invariant I (integrity constraint).• Each transaction is designed to preserve I• If transactions are executed simultaneously, then

they may interfere and invalidate I. • The task of concurrency control is to preserve I.

Banking SystemDeposit(amount, account)

{ x = db.account; x = x + amount; db.account = x; }

Transaction 1: Deposit $50 in Acc1Transaction 2: Deposit $70 in Acc2

Possible interleavings

T1.x = db.Acc1; T1.x = T1.x + 50 T2.x = db.Acc1; T2.x = T2.x + 70; db.Acc1 = T1.x db.Acc1 = T2.x

Concurrency Control

• General organization of managers for handling transactions.

Concurrency Control• General organization of

managers for handling distributed transactions.

• A schedule is any execution of a set of transaction operations

• Two schedules T1 and T2 are equivalent if - all read operations return the same value in both schedules - the final database state is the same in both

schedules

T1: r1(x)0 w1(x)1T2: r2(y)0 r2(x)1 w2(y)2

T1: r1(x)0 w1(x)1

T1: r1(x)0 w1(x)1

T2: r2(y)0 r2(x)1 w2(y)2

T2: r2(y)0 r2(x)1 w2(y)2

T1: r1(y)0 w1(x)1

T2: w2(y)1 r2(x)1 w2(y)2

T1: r1(y)1 w1(x)1

T2: w2(y)1 r2(x)1 w2(y)2

T1: r1(y)0 w1(x)1

T2: w2(y)1 r2(x)0 w2(y)2

• A serial schedule is a schedule in which transactions execute one at a time.

• We know that a serial schedule preserves IC of the database

• A concurrency control algorithm can restricts the execution so that all schedules are serial.

• A CC ensures that all schedules are equivalent to some serial schedule

• A schedule that is equivalent to a serial schedule is called serializable

Untyped Concurrency control• Assumes that all transactions with intersecting

read and write sets interfere with one another.• How can we determine whether a schedule is

serializable• Let T1,…,Tn be a set of transactions• Define a graph G with transactions as nodes• There is an edge from Ti to Tj if - there exists a read rj(x) which reads from wi(x) - there exists a read ri(x) that occurs before wj(x) - there exists a write wi(x) that occurs before wj(x)

• A graph is serializable if the graph is acyclic

Two-phase Locking

• Obtain a read or write lock before reading or writing a variable respectively.

• rl(x): read lock operation• ul(x): unlock operation• wl(x): write lock operation

• Locking rules: - two read locks can be given at the same time; read and write lock must be exclusive * conflict table

• Simple locking does not ensure serializability

T1: r1(y)1 w1(x)1

T2: w2(y)1 r2(x)1 w2(y)2

T1: rl(y) r1(y)1 ul(y) wl(x) w1(x)1 ul(x)

T2: wl(y) w2(y)1 ul(y) rl(x) r2(x)1 ul(x) wl(y) w2(y)2 ul(y)

Two phase locking rule

• Locking phase: acquire all locks • Unlocking phase: release all locks• Two-phase locking ensures serializability• It is prone to deadlocks

Two-Phase Locking (1)

• Two-phase locking.

Two-Phase Locking (2)

• Strict two-phase locking.

Writeahead Log

• a) A transaction• b) – d) The log before each statement is executed

x = 0;y = 0;BEGIN_TRANSACTION; x = x + 1; y = y + 2 x = y * y;END_TRANSACTION; (a)

Log

[x = 0 / 1]

(b)

Log

[x = 0 / 1][y = 0/2]

(c)

Log

[x = 0 / 1][y = 0/2][x = 1/4]

(d)

Graph based protocols

• Impose a partial ordering on data items• If d1 d2, then any transaction accessing

both d1 and d2 must first access d1 before d2.

Tree protocol

• Only exclusive locks are allowed• First item to be locked can be any one• Next, a data item can be locked only if the

parent is already locked• Data items can be unlocked at any time• A data item cannot be relocked by a

transaction.

Semantics-based concurrency control

• If transactions T1 and T2 do not interfere then they can be executed concurrently.

• Two operations op1 and op2 do not conflict if they commute (that is, op1; op2 is the same as op2; op1)

Predicate Locking

• Each transaction specifies a predicate as a lock.

• A new transaction can execute if it does not interfere with existing predicate locks