12.13.Fault Tolerance - Transactions

Question 1

What is a characteristic that distinguishes distributed systems from single-machine systems?

Answer 1

Partial failure

Question 2

What is the goal when partial failure occurs?

Answer 2

Tolerate faults

Question 3

What is being fault tolerant related to?

Answer 3

Dependability

Question 4

What is dependability>

Answer 4

the trustworthiness of a computing system which allows resilliance to be justifiably placed on the service it delivers

Question 5

What are the requirements for Dependability?

Answer 5

1. Availability
2. Reliability
3. Maintainability
4. Safety

Question 6

What does Safety mean?

Answer 6

If and when FAILURES occur the CONSEQUENCES are not catastrophic for the system

Question 7

What does Availability mean?

Answer 7

the probability that the system operates correctly at ANY GIVEN MOMENT

Question 8

What does Reliability mean?

Answer 8

LENGTH OF TIME that it can run continuously without failure

Question 9

What does Maintainability mean?

Answer 9

how EASILY a failed system can be REPAIRED

Question 10

Different types of failures?

Answer 10

1. Crash
2. Omission
3. Response
4. Timing
5. Arbitrary (Byzanitine)

Question 11

What is a technique for failure masking?

Answer 11

Redundancy

Question 12

How many types of redundancy are there?

Answer 12

• Physical
• Information (send extra bits to allow for recovery if need be)
• Time (repeat action if need be)

Question 13

What is one of our most important considerations in failure masking?

Answer 13

Making sure that a failure won’t leave the system in an inconsistent state

Question 14

How is avoiding leaving the system in an inconsistent state achieved?

Answer 14

1.Atomic operations!

“The sequence of operations must execute as an ATOMIC operation”

Question 15

When do concurrent executions not interfere with each other?

Answer 15

If their execution is equivalent to a serial one (they don’t interleave)

Question 16

What does the property of Isolation in distributed systems refer to?

Answer 16

Isolated excecution (concurrent applications)

Question 17

What should the distributed application not violate in order to achieve Consistency?

Answer 17

Database’s integrity constraints

Question 18

What does durability mean in distributed systems?

Answer 18

Changes to the database are persistent

Question 19

What concept allows for the reinforcement of the ACID propertires?

Answer 19

transactions

Question 20

What is a transaction?

Answer 20

A set of operations that is either fully committed or aborted as a whole. If aborted no operation in the set is executed.

Question 21

What algorithms in the implementation of transactions allow for ISOLATION?

Answer 21

Concurrency control

Question 22

What do concurrency control algorithms do to ensure ISOLATION?

Answer 22

Ensure execution is equivalent to “serial” execution

Question 23

What algorithms in the implementation of transactions allow for DURABILITY?

Answer 23

Recovery algorithms

Question 24

What do recovery algorithms do to ensure DURABILITY?

Answer 24

• replay actions of committed transactions

- undo effects of aborted transactions

Question 25

Two ways to improve concurrency control with locking?

Answer 25

- Optimistic concurrency control (transaction executed normal, checked at commit, aborted if problematic) - Timestamp ordering (operations in transactions validated when carried out)

Question 26

Two ways to do recovery when transaction needs to be aborted?

Answer 26

- Backwards (through state checkpoints->Previously correct state) - Forwards (correct new state)

Question 27

What problem arises when trying to make transactions where more than one server is involved?

Answer 27

the distributed commit problem(ATOMICITY) | Either all servers commit or all abort

Question 28

Protocol to support distributed transactions? (more than one server)

Answer 28

1. pick coordinator 2. client communicates transaction to coordinator 2. One or Two phase commit (coordinator communicates abort of transaction to servers)

Question 29

What is the difference between a one phase commit and a two phase commit when dealing with distributed transactions?

Answer 29

Two phase commit involves the servers being able to Accept and Execute the commit (rather than just receive the command)

Question 30

What are the drawbacks of the 2-phase commit?

Answer 30

- Coordinator fail (Three-phase commit and multicast?) - participants must trust coordinator - tranaction must be short - distributed deadlock risk

Question 31

How can deadlock be resolved?

Answer 31

By aborting one of the transactions

Question 32

When does deadlock occur?

Answer 32

When there is a cycle in the wait- for graph of transactions for locks

Question 33

What complicates detecting deadlock in a distributed system?

Answer 33

- Locks are held on different servers | - loop in the entire wait-for graph will not be apparent to any one server

Question 34

One (bad) way of detecting distributed deadlock?

Answer 34

Coordinator stores entire wait-for graph (centra point of failure)

Question 35

What is a better way to detect distributed deadlock?

Answer 35

Edge chasing (Path pushing)