Unit 6 The database tier: Handling transactions and concurrency

Question 1

Compare the concept of a transaction with that of a coarse-grained atomic action. What are their differences and similarities?

Answer 1

(Part 3) A tranasction and a coarses-grained atomic action both consist of several operations, but these are treated as belonging together, and so are both atomic. They give the appearance of indivisibility.

The different is that a transation also has the property of failure atomicity.

• if there is afailure part-way through the execution of the transaction, all the partly done operations are undone, so that it is as if nothing has happened
• if the transaction completes, the results of the transaction are made durable (stored in persistent storage)

Question 2

What are the desirable properties for transactions?

Answer 2

(Part 3)There are four desirable properties: (ACID) atomicity, consistency, isolation and durability.

Question 3

What are the transaction states in the transaction state model?

Answer 3

(Part 3) The transaction model specified the following transaction states: Active, Failed, Partially Commited, Committed and Aborted.

Question 4

When a transaction is in the Partially Committed state, it has not fully completed. What are the possible next states? Describe the circumstances under which it would progress to each of them.

Answer 4

(Part 3) When a transaction is Partially Committed it can progress to the Committed state in the case where all the data involve is written to disk and made durable. If there is a problem in making the data durable the transaction is said to have Failed. The effects of the transaction are then undone (rolled back) and only then is the transaction completely finished when it enters the Aborted state.

Question 5

What is the role of a TP system?

Answer 5

(Part 3) The task of the TP system is to manage the correct and efficient execution of transactions.

Question 6

What is a serial schedule?

Answer 6

(Part 4) A serial schedule is one in which transactions execute strictly one after the other, without any interleaving of operations.

Question 7

What is the definition of serialisability?

Answer 7

(Part 4) A schedule for a group of transactions is serialisable if, and only if, it produces the same results as if the transactions had executed in some serial order.

Question 8

When executing a number of transactions, several serial schedules may be possible. These serial schedules do not necessarily all give the same final results for the data objects involved. Is this problematic?

Answer 8

(Part 4) This is not problematic, as the main concern is whether the data objects are in a consistent state.

Question 9

What are conflicting operations?

Answer 9

(Part 4) Operations conflict if the order in which they are performed affects the result.

Question 10

How is the notion of conflicting operations used to determine whether a schedule is serialisable?

Answer 10

(Part 4) The notion of conflicting operations allows us to determine in what order transactions execute a pair of conflicting operations. The, if for all the pairs of conflicting operations in a schedule the order of execution by transactions is the same, we know that the schedule is equivalent to a serial schedule.

Question 11

Precedence graphs can be used as a notation for transactions and the conflicting operations of those transactions. Under which circumstances could such graphs contain a cycle?

Answer 11

(Part 4) A predence graph contains a cycle if the schedule it represents is not serialisable.

Question 12

What are the two phases of two-phase locking?

Answer 12

(Part 5) The two phases are acquire and release.

In the acquire phase, the transaction gradually gets hold of all the locks it needs, as and when it needs them, but does not release any locks.

In release, the transaction acquires no more locks but lets go of the ones it holds when it has completed work on the locked objects.

Question 13

What is a shared lock and why might we want to use shared locks?

Answer 13

(Part 5) A shared lock is useful when non-conflicting operations are using the same object, as so it is safe for both to use the object. If a shared lock can be used, it increases concurrency.

Question 14

What is a simple alternative to deadlock detection?

Answer 14

(Part 5) Timeouts.

Question 15

What are cascading aborts and what causes them?

Answer 15

(Part 5) Cascading aborts occur when one transaction has to abort and this leads to other transactions having to abort. This is bad for the performance of a system, as much work may have to be repeated.

Question 16

What are the advantages and disadvantages of TSO?

Answer 16

(Part 5) The advantages are:

• simple and efficient implementation
• concurrency control data is held with individual objects and not centrally
• objects are not locked for longer than the duration of an operation and so circular waits and hence deadlocks cannot occur

The main disadvantages of TSO are:

• the loss of flexibility, in that it imposes one serial schedule and excludes other possibilities
• it is more susceptible to transaction aborts than other approaches, due to lack of isolation, and may possibly lead to cascading aborts

Question 17

Why is the optimistic concurrency control approach so named?

Answer 17

(Part 5) This approach is called optimistic because it expects success rather than failure. That is it works, confident that things will turn out operating succesfully and only performs a final check at the very end.

Question 18

What are the distinctions between volatile memory, persistent storage and stable storage?

Answer 18

(Part 6) Volatile memory holds data that would be lost if the device is switched off. Data stored in persisten storage remains even when the device is switched off, or the prgram that created the data stops executing. Stable storage is even more secure, because data is stored in several places; if a disk fails in one location, usually the data survives in a different location.

Question 19

For crash resilience, what must be done when a transaction aborts?

Answer 19

(Part 6) For a transaction, when a crash occurs, it must be as though the transaction had never been invoked. Therefore all the effects of an aborted transaction must be undone before it can be restarted.

Question 20

Which of the ACID properties of transactions are addressed through the provision of crash resilience?

Answer 20

(Part 6) Atomicity and durability. Transactions have to be all or nothing (failure atomic) and therefore crash-resilience mechanisms are put in place to ensure that a half-completed transaction can be undone should a crash occur.

A crash-resilience mechanism also ensures that, if the transaction had committed, its results are definitely stored in persistent store - i.e. durability.

Question 21

What is meant by rolling back?

Answer 21

(Part 6) Rolling back means restoring the persistent store to the state that existed before the start of a transaction that was aborted due to a system crash.

Question 22

In logging, what might happen if log information is written to the persistent store after the data updated by the atomic operation is written there?

Answer 22

(Part 6) In logging, the information in the log is used to roll back the persistent store to its state at the start of the transaction in which the crash occured, by using the old values recorded in the log. If a crash occurs after the data updated by the transaction is written to the permanent store but before the log information is written (it is not a write-ahead log) then there will be no record of the original state to facilitate rollback.

Question 23

What is shadowing? What essential feature is required to ensure that shadowing is implemented successfully?

Answer 23

(Part 6) Shadowing is where the results of a transaction are built up in a structure that mirrors part of the persistent store but the persistent store is not updated until the transaction commits. Once the transaction has committed, the updated copy replaces the relevant part of the persistent store.

The essential feature of shadowing is that the replacement of the relevant part of the persistent store by the updated copy should occur in a single operation (e.g. by setting a single pointer). This avoids problems that would occur if a crash happened during the replacement.

Question 24

What does the term “relation” refer to in relational database models?

Answer 24

(Part 7) Relation refers to the logical grouping of the data, which is often represented as a table, with rows and columns. Each column contains data of a certain type, and each row, also known as a record, contains the details for an object in the table. The relation is the table as a whole.

Question 25

What are the core features of a DBMS?

Answer 25

(Part 7) The core features of a DBMS are:

• a modelling language to model the data
• data structures for storage of data
• a query language to retrieve and update data
• a transaction concurrency control mechanism

Question 26

What is the purpose of JDBC?

Answer 26

(Part 7) The purpose of JDBC is to provide a standard interface for communication with a large number of different databases.

Question 27

What are the steps involved in setting up a connection between a Java application and a database, when using JDBC?

Answer 27

(Part 7) The steps involved in setting up the communication are:

1. the application causes the database driver to be loaded
2. the application requests a connection for a particular database, and the driver manager requests all registered drivers to respond with a connection
3. the driver manager returns a connection to the application
4. communication through the JDBC connection to the database

Question 28

Why is a DBMS not able to offer concurrency control for distributed transactions?

Answer 28

(Part 7) Each DBMS has one resource manager. A distributed transaction involves objects residing in several locations held in several databases, and will require more than one resource manager to manage the concurrency control.

Question 29

What is the default setting for transaction processing when executing SQL statements?

Answer 29

(Part 7) The default setting of the Connection object is that each SQL statement is treated as a transaction

Question 30

How can the default setting be altered?

Answer 30

(Part 7) The default setting can be altered with the command setAutoCommit(false), in which case all the statements after that are treated as forming part of one transaction, which will finish as soon as the commit or the rollback command is executed.

Question 31

Explain how we can deal with possible SQLExceptions occuring during the execution of SWL statements arriving from a Java application?

Answer 31

(Part 7) The SQL statements that form one transaction should be grouped in a try-catch clause. If an exception occurs during one of the statements being executed, this can be caught, and the transaction can then be rolled back using the rollback method.

Question 32

What is the point of having transaction isolation levels?

Answer 32

(Part 7) For some applications it may be appropriate to fine-tune the concurrency control, based on allowing certain types of read to happen.

Question 33

Explain the different forms of read: dirty read, non-repeatable read and phantom read.

Answer 33

(Part 7) A dirty read occurs if a transaction reads values that are written by another transaction that hasn’t commited yet.

A non-repeatable read occurs if a transaction reads the same object twice during execution and finds a different value the second time, although the transaction has not changed the value in the meantime.

A phantom read occurs when a transaction re-executes a query, returning a set of data that satisfies the condition in the query, and finds that the set of data has changed as a result of another recently commited transaction.

Question 34

What are two strategies for the replication of data?

Answer 34

(Part 8) Two strategies for the replication of data are synchronous and asynchronous replication.

In synchronous replication all replicas of a data item are updated at the same time, when a data item is being updated. A transaction cannot commit unless all replicas will be updated.

In asynchronous replication the replicas are updated after the source has been updated.

Question 35

What are the differences between the three forms of ownership for the replication of data?

Answer 35

(Part 8) Three forms of ownership for the replication of data are:

• master-slave ownership - only the master site is permitted to make updates
• workflow ownership - permission to update the data moves along a certain path
• update anywhere ownership - permission is equally shared between peers

Question 36

What do we mean by transparent management of a distributed DBMS? How does this apply to fragmentation and replication?

Answer 36

(Part 8) Transparent management is an important aim of a distributed DBMS, meaning that to the user of such a system it should appear as if they are dealing with just one database. The management of the distribution involves hiding the low-level details of where everything is stored exactly (fragmentation) and whether there are several copies of each data item (replication).

Question 37

What does the 2PC protocol aim to achieve?

Answer 37

(Part 8) The aim of 2PC is that the committing of a distributed transaction is done atomically - that is, either the entire transaction has taken place and its effects are made durable, or it is as if nothing has happened.