Data Parallelism:
Database parallelism in a data warehouse means splitting data processing tasks
across multiple processors or machines to handle large datasets and complex
queries faster and more efficiently.
Types of Database Parallelism:
Parallelism in databases speeds up query execution by using more resources and
manages larger workloads without delays by increasing parallel processing.
It is implemented using architectures like shared-memory, shared-disk, shared-
nothing, and hierarchical structures.
(a)Horizontal Parallelism:
Horizontal parallelism in a data warehouse splits data rows across nodes to process the
same task simultaneously, boosting performance.
(b)Vertical Parallelism:
Vertical parallelism in a data warehouse runs different tasks, like scanning or sorting,
simultaneously to improve efficiency.
Intraquery Parallelism:
• Defines execution of a single query in parallel on multiple processors and
disks.
• Essential for speeding up long-running queries.
• DBMS vendors use intraquery parallelism to improve performance.
• Decomposes serial SQL query into lower-level operations like scan, join, sort,
and aggregation.
• Lower-level operations are executed concurrently in parallel.
Interquery Parallelism:
• Interquery parallelism allows multiple queries or transactions to execute in
parallel.
• Database vendors use parallel hardware architectures to handle large client
requests efficiently.
• Successful implementation on SMP systems increases throughput and
supports more concurrent users.
Shared Disk Architecture:
• Implements shared ownership of the entire database between RDBMS
servers.
• Each server can read, write, update, and delete information from the same
shared database.
• DLM components can be found in hardware, operating system, and separate
software layer.
• Reduces performance bottlenecks from data skew and increases system
availability.
• Eliminates memory access bottleneck of large SMP systems and reduces
DBMS dependency on data partitioning.
Shared-Memory Architecture:
Shared-Memory RDBMS Implementation
• Traditional RDBMS implementation on SMP hardware.
• Simple to implement, but faces scalability limitations.
• Single RDBMS server can apply all processors, access all memory, and the
entire database.
• Multiple database components communicate via shared memory.
• All processors have access to all data partitioned across local disks.
Shared-Nothing Architecture:
• Data partitioned across all disks.
• DBMS partitioned across multiple co-servers.
• Each node owns its disk and database partition.
• Parallelizes SQL query execution across multiple processing nodes.
• Each processor communicates with other processors via interconnection
network.
• Optimized for Multi-Process-Performer-Node (MPP) and cluster systems.
• Offers near-linear scalability, with each node capable of being a powerful
SMP system.
Application of Data Parallelism:
Query Processing: Parallel execution of queries on large datasets to improve
performance.
Data Aggregation: Distributing data across nodes to perform aggregations
simultaneously.
ETL Processes: Dividing ETL tasks (Extract, Transform, Load) into smaller,
parallelizable units.
Indexing and Searching: Splitting indexing tasks to quickly process large volumes of
data.
Advantages:
1. Improved Performance: Faster query execution by processing data in parallel.
2. Scalability: Efficiently handles large volumes of data as workloads can be distributed.
3. Better Resource Utilization: Makes full use of available CPU, memory, and disk
resources.
4. Reduced Processing Time: Divides tasks into smaller units, significantly reducing
overall processing time.
Disadvantages:
1. Complexity in Data Distribution: Proper partitioning and managing data across
nodes can be complex.
2. Overhead for Small Tasks: For small datasets, the overhead of managing parallelism
may outweigh the benefits.
3. Data Skew Issues: Uneven data distribution can lead to performance bottlenecks.
4. Resource Contention: Multiple processes may compete for limited resources,
potentially causing delays.
