Presentation is loading. Please wait.

Presentation is loading. Please wait.

CS 542: Topics in Distributed Systems Transactions and Concurrency Control.

Similar presentations


Presentation on theme: "CS 542: Topics in Distributed Systems Transactions and Concurrency Control."— Presentation transcript:

1 CS 542: Topics in Distributed Systems Transactions and Concurrency Control

2 Banking transaction for a customer (e.g., at ATM or browser) Transfer $100 from saving to checking account; Transfer $200 from money-market to checking account; Withdraw $400 from checking account. Transaction (invoked at client): /* Every step is an RPC */ 1. savings.withdraw(100) /* includes verification */ 2. checking.deposit(100) /* depends on success of 1 */ 3. mnymkt.withdraw(200) /* includes verification */ 4. checking. deposit(200) /* depends on success of 3 */ 5. checking.withdraw(400) /* includes verification */ 6. dispense(400) 7. commit Transactions Client Server Transaction

3 Bank Server: Coordinator Interface  All the following are RPCs from a client to the server  Transaction calls that can be made at a client, and return values from the server: openTransaction() -> trans; starts a new transaction and delivers a unique transaction identifier (TID) trans. This TID will be used in the other operations in the transaction. closeTransaction(trans) -> (commit, abort); ends a transaction: a commit return value indicates that the transaction has committed; an abort return value indicates that it has aborted. abortTransaction(trans); aborts the transaction.  TID can be passed implicitly (for other operations between open and close) with CORBA Transactions can be implemented using RPCs/RMIs!

4 Bank Server: Account, Branch interfaces deposit(amount) deposit amount in the account withdraw(amount) withdraw amount from the account getBalance() -> amount return the balance of the account setBalance(amount) set the balance of the account to amount create(name) -> account create a new account with a given name lookup(name) -> account return a reference to the account with the given name branchTotal() -> amount return the total of all the balances at the branch Operations of the Branch interface Operations of the Account interface

5 Transaction  Sequence of operations that forms a single step, transforming the server data from one consistent state to another.  All or nothing principle: a transaction either completes successfully, and the effects are recorded in the objects, or it has no effect at all. (even with multiple clients, or crashes)  A transactions is indivisible (atomic) from the point of view of other transactions  No access to intermediate results/states of other transactions  Free from interference by operations of other transactions But…  Transactions could run concurrently, i.e., with multiple clients  Transactions may be distributed, i.e., across multiple servers

6 Transaction: 1. savings.deduct(100) 2. checking.add(100) 3. mnymkt.deduct(200) 4. checking.add(200) 5. checking.deduct(400) 6. dispense(400) 7. commit Transaction Failure Modes A failure at these points means the customer loses money; we need to restore old state A failure at these points does not cause lost money, but old steps cannot be repeated This is the point of no return A failure after the commit point (ATM crashes) needs corrective action; no undoing possible.

7 Transactions in Traditional Databases (ACID)  Atomicity: All or nothing  Consistency: if the server starts in a consistent state, the transaction ends the server in a consistent state.  Isolation: Each transaction must be performed without interference from other transactions, i.e., the non-final effects of a transaction must not be visible to other transactions.  Durability: After a transaction has completed successfully, all its effects are saved in permanent storage.  Atomicity: store tentative object updates (for later undo/redo) – many different ways of doing this  Durability: store entire results of transactions (all updated objects) to recover from permanent server crashes.

8 Concurrent Transactions:Lost Update Problem  One transaction causes loss of info. for another: consider three account objects Transaction T1Transaction T2 balance = b.getBalance() b.setBalance(balance*1.1) a.withdraw(balance* 0.1) c.withdraw(balance*0.1) T1/T2’s update on the shared object, “b”, is lost a: b:c: 280 c: 80 a: 220 b: 220 b:

9 Conc. Trans.: Inconsistent Retrieval Prob.  Partial, incomplete results of one transaction are retrieved by another transaction. Transaction T1Transaction T2 a.withdraw(100) total = a.getBalance() total = total + b.getBalance b.deposit(100) total = total + c.getBalance T1’s partial result is used by T2, giving the wrong result for T a: b: 00 a: c: total 300 b:

10  An interleaving of the operations of 2 or more transactions is said to be serially equivalent if the combined effect is the same as if these transactions had been performed sequentially (in some order). Transaction T1 Transaction T2 balance = b.getBalance() b.setBalance(balance*1.1) balance = b.getBalance() b.setBalance(balance*1.1) a.withdraw(balance* 0.1) c.withdraw(balance*0.1) Concurrency Control: “Serial Equivalence” a: b:c: 278 c: a: 242 b: == T1 (complete) followed by T2 (complete)

11  The effect of an operation refers to  The value of an object set by a write operation  The result returned by a read operation.  Two operations are said to be conflicting operations, if their combined effect depends on the order they are executed, e.g., read-write, write-read, write-write (all on same variables). NOT read-read, NOT on different variables.  Two transactions are serially equivalent if and only if all pairs of conflicting operations (pair containing one operation from each transaction) are executed in the same order (transaction order) for all objects (data) they both access.  Why? Can start from original operation sequence and swap the order of non-conflicting operations to obtain a series of operations where one transaction finishes completely before the second transaction starts  Why is the above result important? Because: Serial equivalence is the basis for concurrency control protocols for transactions. Checking Serial Equivalence – Conflicting Operations

12 Read and Write Operation Conflict Rules Operations of different transactions ConflictReason read NoBecause the effect of a pair ofread operations does not depend on the order in which they are executed readwriteYesBecause the effect of aread and awrite operation depends on the order of their execution write YesBecause the effect of a pair ofwrite operations depends on the order of their execution

13  An interleaving of the operations of 2 or more transactions is said to be serially equivalent if the combined effect is the same as if these transactions had been performed sequentially (in some order). Transaction T1 Transaction T2 balance = b.getBalance() b.setBalance(balance*1.1) balance = b.getBalance() b.setBalance(balance*1.1) a.withdraw(balance* 0.1) c.withdraw(balance*0.1) Concurrency Control: “Serial Equivalence” a: b:c: 278 c: a: 242 b: == T1 (complete) followed by T2 (complete) Pairs of Conflicting Operations

14 Conflicting Operators Example Transaction T1 Transaction T2 x= a.read() a.write(20) y = b.read() b.write(30) b.write(x) z = a.read() x= a.read() a.write(20) z = a.read() b.write(x) y = b.read() b.write(30) Serially equivalent interleaving of operations (why?) Conflicting Ops. Non- serially equivalent interleaving of operations

15 Inconsistent Retrieval Prob  Partial, incomplete results of one transaction are retrieved by another transaction. Transaction T1Transaction T2 a.withdraw(100) total = a.getBalance() total = total + b.getBalance b.deposit(100) total = total + c.getBalance T1’s partial result is used by T2, giving the wrong result for T a: b: 00 a: c: total 300 b:

16 A Serially Equivalent Interleaving of T1 and T2 TransactionT1: a.withdraw(100); b.deposit(100) TransactionT2: aBranch.branchTotal() a.withdraw(100); $100 b.deposit(100) $300 total = a.getBalance() $100 total = total+b.getBalance() $400 total = total+c.getBalance()...

17  How can we prevent isolation from being violated?  Concurrent operations must be consistent:  If trans.T has executed a read operation on object A, a concurrent trans. U must not write to A until T commits or aborts.  If trans. T has executed a write operation on object A, a concurrent U must not read or write to A until T commits or aborts.  How to implement this? Implementing Concurrent Transactions

18 Concurrency control Lost update –3 accounts (A, B, C) »with balances 100, 200, 300 –T1 transfers from A to B, for 10% increase –T2 transfers from C to B, for 10% increase –Both T1, T2 read balance of B (200) –T1 overwrites the update by T2 »Without seeing it Transactions should not read a “stale” value & use it in computing a new value

19 Concurrency control Inconsistent retrievals –T1: transfers 10% of account A to account B –T2: computes sum of account balances –T2 computes sum before T1 updates B Update transactions should not interfere with retrievals. In general: Transactions should not violate operation conflict rules.

20 Concurrency control Serial equivalence criterion for correct concurrent execution T1 serially equivalent with T2 iff: All pairs of conflicting operations of the two transactions are executed in the same order at all objects that both transactions access. 3 approaches to CC: - Locking - Optimistic CC - Timestamp ordering Tx’s wait for one another OR: Restart Tx’s after conflicts have been detected

21 Recoverability from aborts Servers must prevent a aborting Tx from affecting other concurrent Tx’s. –Dirty reads: »T2 sees result update by T1 on account A »T2 performs its own update on A & then commits. »T1 aborts -> T2 has seen a “transient” value T2 is not recoverable »If T2 delays its commit until T1’s outcome is resolved: Abort(T1) -> Abort(T2) However, if T3 has seen results of T2: –Abort(T2) -> Abort(T3) ! »Cascading aborts Tx’s should only read values written by committed Tx’s

22 Recoverability from aborts Premature writes: –Assume server implements abort by maintaining the “before” image of all update operations »T1 & T2 both updates account A »T1 completes its work before T2 »If T1 commits & T2 aborts, the balance of A is correct »If T1 aborts & T2 commits, the “before” image that is restored corresponds to the balance of A before T2 »If both T1 & T2 abort, the “before” image that is restored corresponds to the balance of A as set by T1 Tx’s should be delayed until earlier Tx’s that update the Same objects have been either committed or aborted.

23 Recoverability from aborts Tx’s should delay both their reads & updates in order to avoid interference –Strict execution -> enforce isolation Servers should maintain tentative versions of objects in volatile memory Tx’s should be delayed until earlier Tx’s that update the Same objects have been either committed or aborted.

24 Concurrency Control: Locks Transactions: –Must be scheduled so that their effect on shared data is serially equivalent –Two types of approach »Pessimistic  If something can go wrong, it will Operations are synchronized before they are carried out »Optimistic  In general, nothing will go wrong Operations are carried out, synchronization at the end of the transaction –Locks (pessimistic) »can be used to ensuring serializability »lock(x), unlock(x )

25 Locks: Basics Oldest and most widely used CC algorithm A process before read/write  requests the scheduler to grant a lock Upon finishing read/write  the lock is released In order to ensure serialized transaction Two Phase Locking (2PL) is used

26 Locking How Locks prevent consistency problems –Lost update and inconsistent retrieval: –Causes: »are caused by the conflict between r i (x) and w j (x) »two transactions read a value and use it to compute new value –Prevention: »delay the reads of later transactions until the earlier ones have completed Disadvantage of Locking –Deadlocks

27 2PL Strict 2PL avoids Cascading Aborts –A situation where a committed transaction has to be undone because it saw a file it shouldn’t have seen. Problems of Locking –Deadlocks –Livelocks »A transaction can’t proceed for an indefinite amount of time while other transactions continue normally. It happens due to unfair locking. –Lock overhead »If the system doesn’t allow shared access--wastage of resources –Avoidance of Cascading Aborts may be costly »Strict 2PL in fact, reduces the effect of concurrency

28  Exclusive Locks Transaction T1 Transaction T2 OpenTransaction() balance = b.getBalance() OpenTransaction() balance = b.getBalance() b.setBalance(balance*1.1) a.withdraw(balance* 0.1) CloseTransaction() b.setBalance(balance*1.1) c.withdraw(balance*0.1) CloseTransaction() Example: Concurrent Transactions Lock B Lock A UnLock B UnLock A Lock C UnLock B UnLock C … WAIT on B Lock B …

29  Transaction managers (on server side) set locks on objects they need. A concurrent trans. cannot access locked objects.  Two phase locking:  In the first (growing) phase of the transaction, new locks are only acquired, and in the second (shrinking) phase, locks are only released.  A transaction is not allowed acquire any new locks, once it has released any one lock. Basic Locking

30  Strict two phase locking:  Locking on an object is performed only before the first request to read/write that object is about to be applied.  Unlocking is performed by the commit/abort operations of the transaction coordinator.  To prevent dirty reads and premature writes, a transaction waits for another to commit/abort  However, use of separate read and write locks leads to more concurrency than a single exclusive lock – Next slide Basic Locking

31 non-exclusive lock compatibility Lock alreadyLock requested setreadwrite none OK OK read OKWAIT writeWAITWAIT  A read lock is promoted to a write lock when the transaction needs write access to the same object.  A read lock shared with other transactions’ read lock(s) cannot be promoted. Transaction waits for other read locks to be released.  Cannot demote a write lock to read lock during transaction – violates the 2P principle 2P Locking: Non-exclusive lock (per object)

32 Two Phase Locking (2PL) Protocols In 2PL—All lock operations must precede the first unlock operation –Two phases »expanding or growing phase: all locking are done in this phase but no lock release allowed »shrinking phase: all lock release but no lock acquire Growing phase Shrinking phase Time No. of Locks

33  When an operation accesses an object:  if you can, promote a lock (nothing -> read -> write)  Don’t promote the lock if it would result in a conflict with another transaction’s already-existing lock  wait until all shared locks are released, then lock & proceed  When a transaction commits or aborts:  release all locks that were set by the transaction Locking Procedure in Strict-2P Locking

34  Non-exclusive Locks Transaction T1 Transaction T2 OpenTransaction() balance = b.getBalance() OpenTransaction() balance = b.getBalance() b.setBalance(balance*1.1) Commit Example: Concurrent Transactions R-Lock B … Cannot Promote lock on B, Wait Promote lock on B

35  What happens in the example below? Transaction T1 Transaction T2 OpenTransaction() balance = b.getBalance() OpenTransaction() balance = b.getBalance() b.setBalance(balance*1.1) b.setBalance=balance*1.1 Example: Concurrent Transactions R-Lock B … Cannot Promote lock on B, Wait …

36 Deadlock with write locks TransactionT U OperationsLocksOperationsLocks a.deposit(100); write lockA b.deposit(200) write lockB b.withdraw(100) waits forU’sa.withdraw(200);waits forT’s lock onB A T locks A and waits for U to release the lock on B, U on the other hand locks B and waits for T to release the lock on A  Circular hold and wait  Deadlock

37 The corresponding wait-for graph B A Waits for Held by T U U T Waits for

38 Deadlocks  Necessary conditions for deadlocks  Non-shareable resources (exclusive lock modes)  No preemption on locks  Hold & Wait or Circular Wait T U Wait for Held by Wait for A B T U Held by Wait for A B V W... Wait for Held by Hold & Wait Circular Wait

39 Naïve Deadlock Resolution Using Timeout Transaction TTransaction U OperationsLocksOperationsLocks a.deposit(100); write lock A b.deposit(200) write lock B b.withdraw(100) waits for U ’s a.withdraw(200); waits for T’s lock onB A (timeout elapses) T’s lock onA becomes vulnerable, unlockA, abort T a.withdraw(200); write locksA unlockA, B Disadvantages?

40 Strategies to Fight Deadlock  Lock timeout (costly and open to false positives)  Deadlock Prevention: violate one of the necessary conditions for deadlock (from 2 slides ago), e.g., lock all objects before transaction starts, aborting entire transaction if any fails  Deadlock Avoidance: Have transactions declare max resources they will request, but allow them to lock at any time (Banker’s algorithm)  Deadlock Detection: detect cycles in the wait-for graph, and then abort one or more of the transactions in cycle

41 Optimistic Concurrency Control (Kung and Robinson) We have seen locking has some problems OCC based on the following simple idea: –Don’t worry about the conflicts, keep on doing whatever you’re doing, if there’s a problem worry about it later.

42 Optimistic Concurrency Control (Kung and Robinson) Algorithm –Each transaction has the following phases »Working phase Each transaction has a tentative version of each object that it updates Tentative version allows the trans. to abort w/o affecting the object »Validation phase transaction is validated to see if any conflicts with other trans. »Update phase if a trans. is validated all tentative objects are made permanent

43 Optimistic Concurrency Control: Earlier committed transactions WorkingValidationUpdate T 1 T v Transaction being validated T 2 T 3 Later active transactions active 1 2 Validation of transactions

44 Validation Rules TvTiRule writeread1. Ti must not read objects written by Tv readwrite2. Tv must not read objects written by Ti write 3. Ti must not write objects written by Tv and Tv must not write objects written by Ti

45 Validation of Transactions Backward validation of transaction T v boolean valid = true; for (int T i = startTn+1; T i <= finishTn; T i ++){ if (read set of T v intersects write set of T i ) valid = false; } Forward validation of transaction T v boolean valid = true; for (int T id = active1; T id <= activeN; T id ++){ if (write set of T v intersects read set of T id ) valid = false; }

46 Summary Increasing concurrency important because it improves throughput at server Applications are willing to tolerate temporary inconsistency and deadlocks in turn –Need to detect and prevent these Driven and validated by actual application characteristics – mostly-read transactions abound


Download ppt "CS 542: Topics in Distributed Systems Transactions and Concurrency Control."

Similar presentations


Ads by Google