Concurrency Control in Distributed Databases

Gul Sabah Arif

Introduction
Concurrency control is the activity of coordinating concurrent accesses to a database in a multi-user database management system (DBMS). Several problems
1. The lost update problem. 2. The temporary update problem. 3. The incorrect summary problem.

Serializability Theory.

Distributed DD Architecture.
Transaction Manager Data Manager Scheduler

DDBS Architecture

Processing Operation

Scheduling Algorithms
Modify concurrency control schemes for use in distributed environment. There are 3 basic methods for transaction concurrency control.

Locking (two phase locking - 2PL). Timestamp ordering Optimistic Hybrid

Locking Protocols
Majority Protocol
Local lock manager at each site administers lock and unlock requests for data items stored at that site. When a transaction wishes to lock an un replicated data item Q residing at site Si, a message is sent to Si s lock manager.

If Q is locked in an incompatible mode, then the request is delayed until it can be granted. When the lock request can be granted, the lock manager sends a message back to the initiator indicating that the lock request has been granted.

Majority Protocol (Cont.)

In case of replicated data

Benefit

If Q is replicated at n sites, then a lock request message must be sent to more than half of the n sites in which Q is stored. The transaction does not operate on Q until it has obtained a lock on a majority of the replicas of Q. When writing the data item, transaction performs writes on all replicas.
Can be used even when some sites are unavailable Requires 2(n/2 + 1) messages for handling lock requests, and (n/2 + 1) messages for handling unlock requests. Potential for deadlock even with single item - e.g., each of 3 transactions may have locks on 1/3rd of the replicas of a data.

Drawback

Biased Protocol
Local lock manager at each site as in majority protocol, however, requests for shared locks are handled differently than requests for exclusive locks. Shared locks. When a transaction needs to lock data item Q, it simply requests a lock on Q from the lock manager at one site containing a replica of Q. Exclusive locks. When transaction needs to lock data item Q, it requests a lock on Q from the lock manager at all sites containing a replica of Q. Advantage - imposes less overhead on read operations. Disadvantage - additional overhead on writes

2 Phase Locking (2PL)

Centralized 2PL. Primary copy 2PL. Distributed 2PL. Voting 2PL.

Simulation Models for 2PL

Simulation model of centralized 2PL

Simulation model of distributed 2PL

Timestamping
Secure Time-Stamp Based Concurrency Control Protocol For Distributed Databases
Security level is assigned to each transaction and data. common instances of totally ordered security levels are the Top-Secret (TS), Secret (S), Confidential(C), and Unclassified (U).

System Model

N number of sites, where each site Ni is having a secure database, which is a partition of global database scattered on all the N sites. The secure distributed database is defined as a five tuples < Dt ,Tt ,Ts , Sc ,Lv >
Dt is the set of data items, Tr is the set of distributed transactions, Ts is the timestamp Sc is the partially ordered set of security levels Lv is a mapping

Secure Time-Stamp Based Concurrency Control (cont)

Security level Sci is said to dominate security level Sc j if Sc j <= Sci The security policy used enforces the following restrictions: Simple Security Property: A transaction T (subject) is allowed to read a data item (object) x , only if Lv (x) <= Lv (T ) . Restricted Property: A transaction T is allowed to write a data item x only if Lv (x) = Lv (T)

System Architecture
(GTr Global Transaction Manager) is a software module which translates and decomposes the transaction into subtransactions against local schemas, and coordinates the execution of the subtransactions.

1. 2. 3. 4. 5.

GTr Layers
Transaction Interface Authentication Check Layer Security Level Assignment layer Data Manager and Transaction Manager Layer Data Access Tracker(DAT)

Secure Time-Stamp Based Concurrency Control (cont)

Local Transaction Manager (LTr)
Sub transaction interface layer Sub Query Manager Data Administrator Layer Local Database

A Secure Concurrency Control Protocol

If ( RTs(x) > Tsi ) { Abor t ( Si ) ; } El seI f ( WTs(x) > Tsi ) { Ignor e ( Si ) ; }El seI f( Lv (x) == Lv (Si ) ) /* Lv (x)&Lv (Si ) is security level of data item x & transact ion Si */ { Wr itelockTo( x ) ; Execut ion( x ) ; WTs(x) = Tsi ; Update DAT to Tsi ; }Else{ Abor t( Si ) ;/* acces s denied due to secur ity */} Algorithm for read operation on data item x issued by sub transaction Si with Timestamp Tsi : If (WTs(x) > Tsi ){ Abort( Si ); Rollback( Si ); }ElseIf( Lv (x) <= Lv (Si ) ){ ReadlockTo( x ); ExecuteOn( x ); RTs(x) = Tsi ; Update DAT to Tsi ; }Else{Abort( Si ); Rollback( Si );}

Hybrid
Three basic technique and each can be used for rw or ww scheduling or both. Schedulers can be centralized or distributed. Replicated data can be handled in three ways (Do Nothing, Primary Copy, Voting). System R* Use a 2PL scheduler for rw and ww synchronization. The schedulers are distributed at the DM's. Replication is handled by the do nothing approach. Distributed INGRES INGRES uses primary copy for replication.

New Approaches to Concurrency Control

Total Ordering

Total ordering in networking terms describes the property of a network guaranteeing that all messages are delivered in the same order across all destinations. In combination with the concept of transactions, one can make use of this property to ensure that transactions are received in the same order at all sites called the ORDER CC technique.

Algorithm

Each transaction is initiated by sending its reads and write predeclares to the corresponding schedulers as a single atomic action in totally ordered fashion. Each scheduler stores the received operation requests in a FIFO-type queue. If read is at the head of the queue, it is immediately executed. transaction can now issue the write requests in accordance with the previously given predeclares. Upon commit, the committed values are send in non-ordered fashion to the schedulers, which re-place the corresponding predeclare statements in the queue with the received committed writes.

Timestamp Ordering Revisited

Whenever a network layout provides predictability regarding the time at which a message will arrive at its destination, such as interconnection networks, this property can be exploited for concurrency control . The transaction manager initiates a transaction by sending its reads and write predeclares to the corresponding schedulers as a single atomic action. This atomic action is assigned a timestamp t, denoting the time by which all operations will have arrived at their respective schedulers. When a scheduler receives an operation o, it can either wait until time t has arrived . The alternative option is to process o ahead of time t, and causing conflicting operations that arrive afterwards, but with a lower timestamp, to abort.

Algorithm

Conclusion
Performance Comparison
2PL, the standard technique used for centralised DBMSs, proves to perform rather poorly for distributed systems, whereas timestamp ordering based protocols in their various forms seem to provide the best overall performance. In 2PL, and other locking techniques as well, the deadlock prevention or detection in a distributed environment, which is much more complex and costly . Timestamp ordering techniques (TO) avoid deadlocks entirely. Basic TO (BTO) usually shows better overall performance in a distributed environment than 2PL. ORDER outperforms both 2PL and BTO, i.e. low network latency and an efficient implementation of the total ordering algorithm.For high network latencies, ORDER appears to be a rather disadvantageous approach. PREDICT shows basically the same advantages ORDER does.

References
A Secure Time-Stamp Based Concurrency Control Protocol For Distributed Databases Journal of Computer Science 3 (7): 561-565, 2007 Some Models of a Distributed Database Management System with Data Replication", International Conference on Computer Systems and Technologies CompSysTech07.
A Sophisticated introduction to distributed database concurrency control, Harvard University Cambridge,

1990. Database system concepts,from Silberschatz Mcgraw Hill 2001.