THE IMPLEMENTATION OF SYNCHRONOUS LEADERLESS REPLICATION ON POSTGRESQL
Parallelization in relational databases is to coordinate relational database nodes to obtain higher performance. One solution for relational database parallelization is leader-follower replication. However, this solution only has one node that can modify data, which results in a write bottleneck....
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/78301 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | Parallelization in relational databases is to coordinate relational database nodes to
obtain higher performance. One solution for relational database parallelization is
leader-follower replication. However, this solution only has one node that can
modify data, which results in a write bottleneck. Another solution for
implementing relational database parallelization is through synchronous leaderless
replication. With this replication, each node in the cluster can modify data and still
comply with the ACID rules of relational model.
The development of synchronous leaderless replication in PostgreSQL is carried
out by introducing a lock manager node and enabling WAL shipping between
nodes. A lock manager aims to control the concurrent data modification by each
node. The developed WAL shipping mechanism is an adaptation of PostgreSQL’s
WAL shipping for leader-follower replication.
Performance testing of the development replication was compared with the
performance of PostgreSQL's synchronous leader-follower replication. Results
showed an increase in insert operation performance of 45%. The update and
delete operations experienced a decrease in performance of 3% and 1500%
respectively. It was also found that asynchronous replication generally has faster
data modification performance than the developed replication. |
|---|