Parallel and distributed databases

1
Parallel and distributed databases

• R G Chapter 22

2
What is a distributed database?
3
Why distribute a database

• Scalability and performance
• Resilience to failures

Throughput
Data size
X
versus
X
4
Why distribute a database

• Data is already distributed
• Or needs to be distributed
• Data is in multiple systems

5
Why not distribute a database

• You must earn your complexity!
• Communication needed
• Must build a complex infrastructure
• Unpredictable latencies must be masked
• More types of failures
• More components to fail
• Network failures
• Congestion, timeouts
• More complex planning
• Communication cost plus I/O cost
• May have to deal with heterogeneity
• Different types of systems
• Different schemas, possibly incompatible
• Different administrative domains

6
Types of distributed databases
7
The old days mainframes
Definitely not distributed!
8
Client-server
User interaction
Network
Data processing
9
Parallel database
10
Primary/secondary
X
11
Multidatabase
12
How do they work?

• What is shared?
• How to distribute the data?
• How to process the data?
• How to update the data?

13
What is shared?

• Memory

Most modern DBMSs
CPUs
RAM
Disk
14
What is shared?

• Disk

Oracle RAC
RAM
15
What is shared?

• Nothing

Search engines, Teradata
RAM
16
How to distribute the data?
17
How to distribute the data?
Hash partitioning
Range partitioning
(key,value)
(key,value)
Hash()
lt X
gt X
18
How to distribute the data?
19
Query processing

• Intra-operator parallelism
• Inter-operator parallelism

20
Parallel scanning
Result
filter
filter
filter
filter
filter
filter
21
Sorting
22
Sorting
23
Parallel hash join
Hash()
24
Join
25
Semi-join
26
Inter-operator parallelism
27
Updating distributed data

• Synchronous read-any-write-all

Reads are fast
28
Updating distributed data

• Synchronous voting

29
Updating distributed data

• Synchronous voting

Writes tolerant to disconnection
30
Consistency of distributed data

• Should provide ACID

31
Primary/secondary
32
Two-phase commit
PREPARE
COMMIT
PREPARED
PREPARED
33
Two-phase commit
PREPARE
ABORT
PREPARED
ABORT
34
Two-phase commit
PREPARE
ABORT
PREPARED
35
Two-phase commit
X
PREPARE
PREPARED
PREPARED
36
Conclusion

• Parallelism and distribution very useful
• Performance
• Fault tolerance
• Scale
• But complex!
• Rethink lots of aspects of the system
• Must earn the complexity