PolarDB for PostgreSQL 16 delivers significantly higher OLTP throughput than open source PostgreSQL 16 across all workload types—up to 9.6x faster on write-intensive scenarios and up to 4.5x faster on read-intensive scenarios.
How it works
PolarDB for PostgreSQL 16 achieves this performance advantage through four optimizations:
Hardware acceleration — Built on Optane storage cards based on 3DXpoint storage medium, NVMe SSDs, and RoCE (RDMA over Converged Ethernet) networking.
User-mode I/O and network stack — A full I/O and network protocol stack runs entirely in user mode, bypassing kernel overhead to reduce latency.
Polar CSN (Commit Sequence Number) — Replaces native PostgreSQL xid transactions with a monotonically incrementing 64-bit integer, improving OLTP performance under high concurrency.
I/O reduction techniques — Prefetching, pre-extension, and
rel_size_cachereduce I/O operations and improve throughput.
For the full testing methodology, see Performance testing methodology (OLTP).
Test environment
Both systems used identical CPU and memory specifications. Open source PostgreSQL 16 was deployed on Alibaba Cloud ECS (Elastic Compute Service) instances.
| Parameter | PolarDB for PostgreSQL 16 | Open source PostgreSQL 16 |
|---|---|---|
| CPU cores + memory | 8 cores, 64 GB | 8 cores, 64 GB |
| Load-generating ECS instance type | ecs.u1-c1m2.8xlarge | ecs.u1-c1m2.8xlarge |
| Concurrency | 256 | 256 |
| PostgreSQL version | 16.10 | 16.10 |
| Disk performance level | PSL5 | cloud_essd+PL3 |
| Reserved disk space | Auto-scaling, unlimited | Auto-scaling, unlimited |
Kernel parameters for PolarDB for PostgreSQL 16 were aligned with those of open source PostgreSQL 16 for performance testing. All other parameters used default values.
Test workloads
Tests used Sysbench with seven OLTP workload types:
| Category | Scenario | Description |
|---|---|---|
| Write | oltp_insert | Insert rows into a table |
| Write | oltp_update_index | Update indexed tables |
| Write | oltp_update_non_index | Update non-indexed tables |
| Write | oltp_write_only | Mixed insert, update, and delete operations |
| Read | oltp_point_select | Point queries |
| Read | oltp_read_only | Point queries and range queries |
| Read-write mixed | oltp_read_write | Mixed oltp_write_only and oltp_read_only workload |
Metric: TPS (Transactions Per Second) — the number of transactions the database executes per second, measured by successful COMMIT operations.
Test dataset
| Parameter | Value |
|---|---|
| Number of tables | 8 |
| Number of rows | 64,000,000 |
| Total data volume | 128 GB |
Performance results
Write scenarios
| Scenario | PolarDB for PostgreSQL 16 (TPS) | Open source PostgreSQL 16 (TPS) |
|---|---|---|
oltp_insert | 111,618.70 | 91,512.15 |
oltp_update_index | 77,647.89 | 8,129.04 |
oltp_update_non_index | 78,112.42 | 8,420.75 |
oltp_write_only | 24,602.97 | 2,658.71 |
Read scenarios
| Scenario | PolarDB for PostgreSQL 16 (TPS) | Open source PostgreSQL 16 (TPS) |
|---|---|---|
oltp_point_select | 220,782.88 | 48,837.41 |
oltp_read_only | 8,653.15 | 1,049.34 |
Read-write mixed scenario
| Scenario | PolarDB for PostgreSQL 16 (TPS) | Open source PostgreSQL 16 (TPS) |
|---|---|---|
oltp_read_write | 5,073.52 | 649.34 |
Key takeaways
PolarDB for PostgreSQL 16 outperforms open source PostgreSQL 16 across every tested workload:
Write-intensive workloads — Up to 9.6x higher TPS (
oltp_update_index: 77,647 vs. 8,129).Read-intensive workloads — Up to 4.5x higher TPS (
oltp_point_select: 220,782 vs. 48,837).Mixed workloads — 7.8x higher TPS (
oltp_read_write: 5,073 vs. 649).