Parallel database management system (DBMS)
A parallel database management system (DBMS) is a database management system that works on several processors or CPUs and is primarily designed to execute query operations in parallel whenever possible. The parallel DBMS connects several smaller machines to obtain the same throughput as a single large system. Parallelism is utilized to support query speedup, in which extra resources, like processors and drives, are made available. Scale-up is another application of parallelism, in which increasing workloads are managed without increasing response time by raising the degree of parallelism.
There are three types of architectures for the parallel database management system (DBMS), Here I am going to describe them,
Shared Nothing Architecture:
Shared nothing architecture is a distributed database management system. It is also called the shared-nothing architecture because the nodes are not connected to each other.
The nodes are generally independent, horizontally partitioned, and synchronized by messages that are passed between them. Nodes can be added or removed from the system at any time without affecting the other nodes in the system. This type of architecture can be used to create large-scale systems that are capable of handling an enormous amount of data with high availability, resilience, and scalability
Shared Everything/Shared Disk Architecture:
The concept of shared ownership of the complete database between RDBMS servers, each of which is executing on a node of a distributed memory system, is implemented using shared-disk architecture. Each RDBMS server can read, write, update, and remove data from the same shared database, necessitating the need for a distributed lock manager in the system (DLM)
The discs are accessible from all cluster nodes in this architecture. This architecture is capable of quickly adapting to changing workloads. It employs a robust optimization strategy.
Shared-Memory Architecture:
Shared-memory or shared-everything style is the standard technique of building an RDBMS on SMP hardware. It is quite straightforward to construct and has proven highly effective up to the point when it runs into the shared-everything architecture's scalability limitations. The major benefit of this approach is that a single RDBMS server can most likely use all CPUs, access all memory, and access the whole database, giving the client a consistent single system image.
The DBMS assumes that many database components executing SQL queries interact with each other via shared memory in shared-memory SMP systems. All CPUs have access to all data on the local disc, which is partitioned.
Conclusion:
Different types of architecture suits different application the above explanation simply explains their architecture. Shred Disk Architecture is comparatively more expensive than Shared nothing Architecture but it has Dynamic load balancing. In a shared everything environment, all servers have access to the same shared store, and each workload has access to the store as well as all of the grid's computational resources. The Shared Nothing Architecture is focused on boosting performance, and everything in shared is focused on maximizing resource utilization. So it depends on the requirement to wisely choose the architecture for your Parallel DBMS.
reference:
https://www.javatpoint.com/data-warehouse-types-of-database-parallelism
https://www.geeksforgeeks.org/design-of-parallel-databases-dbms/
https://www.geeksforgeeks.org/difference-between-shared-nothing-architecture-and-shared-disk-architecture/