For more information on pgbench transactions, refer to the “What is the ‘Transaction’ actually performed in pgbench?” section of the official pgbench documentation.
Encryption at rest performance benchmark on Database Instances
In December 2024, the Scaleway Managed Databases team ran tests on DB-POP2-4C-16G Instances in a Private Network using pgbench
. Several workloads (including read, oltp and large loads) were simulated with several scale factors (from 1M rows to 200M rows) and concurrent database connections set to 10. They measured the impact of encryption at rest on transactions, per second and latency.
The workloads applied simulate heavy usage scenarios for the purposes of testing. During the beta of encryption at rest, no performance issues were raised by users. If you would like to share your own benchmarks with our team, feel free to reach out to us in the #database channel of the Scaleway Slack community.
The table below compares the different workload and setup combinations and indicates the performance results of each:
Workload Type | Setup | Scale Factor | Performance Impact |
---|---|---|---|
Read Workloads | Standalone | 10 | No impact |
Standalone | 2000 | No real impact with or without encryption at rest | |
HA | 2000 | 20-30% impact, likely due to disk writes being replicated. Refer to this blogpost to learn more about disk writes in read operations. | |
OLTP Workloads | Standalone | Any | 15% performance drop with encryption at rest activated |
HA | 10 | 15% performance drop | |
HA | 2000 | Up to 30% performance drop | |
Large Data Load (29 GB) | Standalone | Any | 10% additional latency |
HA | Any | 40% additional latency due to semi-sync replication and encryption overhead |