- Distributed database:
- Distributed database =
a database whose
data (relations) are
stored on
multiple autonomous sites
- Distributed database =
a database whose
data (relations) are
stored on
multiple autonomous sites
- Minimum requirements to be called
distributed:
- Nodes (sites) are
interconnected by a
computer network
(not a
LAN (local area network))
- Each node (site) maintains a database
- Data stored at different sites are different
- The various databases
(stored at the sites) are
logically related
- Nodes (computers) at the sites can be heterogeneous (= different types of computers)
- Nodes (sites) are
interconnected by a
computer network
(not a
LAN (local area network))
- Example: bank
- A bank has many branches
- Each branch keeps a database of accounts at that branch
- A central branch office keep bank employee records and fee/interest rate information
Graphically:
- Important differences between
a distributed database and
a shared-nothing parallel system:
- System aspect:
- Nodes of distributed DB can be heterogeneous
- Nodes of shared nothing parallel system are homogeneous
- Performance aspect:
- Distributed database systems has higher communication cost !!!!
- System aspect:
- Shared-nothing
parallel system:
- Storage/processing nodes are located in the same room
- Interconnection network used has Gbps transmission rate
- Distributed database:
- Storage/processing nodes can be located in different states or even countries
- Interconnection network used has Mbps transmission rate
- Consequently:
- Dominant
cost in
shared-nothing
parallel system:
- Disk I/O
- Dominant
cost in
Distributed databases:
- Communication !!!
- Dominant
cost in
shared-nothing
parallel system:
- Consider the
accounts of a
bank that are
stored in
different branches:
- Physically speaking:
- There are multiple physical relations stored at different sites
- Logically speaking:
- There is only one logical relation that store information of the bank accounts of the bank
- A logical relation can be
decomposted
into
mulitple physical relations
in
2 ways:
- Horizontal decomposition:
- The tuples of the logical relations are stored over multiple locations
Graphically:
- Vertical decomposition:
- The logical relation is decomposed into multiple relations (reconstructed using join operations)
Graphically:
Note:
- You need to use join operations to obtain the logical relation....
- Horizontal decomposition:
- Recall:
tuple storage in
shared-nothing parallel system
- Tuples are
hashed and stored
at the site with the
hash ID
- Result:
- Duplicate tuples will be
stored at the same site
- Many relational algebra operations can be performed locally without using other fragments
- Duplicate tuples will be
stored at the same site
- Tuples are
hashed and stored
at the site with the
hash ID
- Distributed Database:
- Tuples are stored
in the site of "creation"
- There is no hash function used to re-distribute the tuples !!!
- Tuples are stored
in the site of "creation"
- System failures:
- When some part of a system
fails, the
data stored on the
failed part becomes
inaccessible
Example:
- When some part of a system
fails, the
data stored on the
failed part becomes
inaccessible
- Fault Tolerance:
- Distributed Databases can
often tolerate (a limited number of)
system (e.g., site) failures
- The commonly used method to
achieve
fault tolerance is
data replication:
Every relation is stored on multiple sites
- Distributed Databases can
often tolerate (a limited number of)
system (e.g., site) failures
- Note:
- The different copies must
be consistent
- We must use additional replica control protocols to keep the multiple copies of data consistent
- The different copies must
be consistent