All Products
Search
Document Center

PolarDB:Compute node specifications

Last Updated:Mar 30, 2026

PolarDB for PostgreSQL (Compatible with Oracle) uses dedicated compute node specifications. Each node has exclusive access to its CPU, memory, storage, and I/O resources — no sharing with other nodes on the same server. This isolation improves stability and predictability under load.

Specifications

Node type CPU and memory Maximum storage capacity Maximum connections PSL4 maximum IOPS PSL5 maximum IOPS
polar.o.x4.medium 2 cores, 8 GB memory 50 TB 800 8,000 16,000
polar.o.x8.medium 2 cores, 16 GB memory 100 TB 1,600 8,000 16,000
polar.o.x4.large 4 cores, 16 GB memory 100 TB 1,600 32,000 64,000
polar.o.x8.large 4 cores, 32 GB memory 100 TB 3,200 32,000 64,000
polar.o.x4.xlarge 8 cores, 32 GB memory 100 TB 3,200 50,000 128,000
polar.o.x8.xlarge 8 cores, 64 GB memory 100 TB 3,200 50,000 160,000
polar.o.x4.2xlarge 16 cores, 64 GB memory 100 TB 3,200 64,000 256,000
polar.o.x8.2xlarge 16 cores, 128 GB memory 100 TB 12,800 64,000 256,000
polar.o.x4.4xlarge 32 cores, 128 GB memory 100 TB 12,800 80,000 256,000
polar.o.x8.4xlarge 32 cores, 256 GB memory 300 TB 25,600 80,000 384,000
polar.o.x4.6xlarge 48 cores, 192 GB memory 100 TB 12,800 100,000 256,000
polar.o.x8.6xlarge 48 cores, 384 GB memory 300 TB 25,600 100,000 384,000
polar.o.x4.8xlarge 64 cores, 256 GB memory 300 TB 25,600 120,000 384,000
polar.o.x8.8xlarge 64 cores, 512 GB memory 500 TB 36,000 120,000 409,600
polar.o.x8.12xlarge 88 cores, 710 GB memory 500 TB 36,000 150,000 512,000
polar.o.x8.15xlarge 120 cores, 920 GB memory 500 TB 36,000 150,000 512,000

Node type naming: The x4 and x8 suffix indicates the memory-to-CPU ratio. x4 nodes provide 4 GB of memory per core; x8 nodes provide 8 GB per core. Choose x8 for memory-intensive workloads.

Important

The PSL4 IOPS values in the table are maximums under ideal conditions. Actual IOPS for PSL4 storage depends on your storage capacity and may be significantly lower. PSL5 IOPS is not storage-capacity dependent — the values in the table reflect what you get regardless of storage size. See How PSL4 IOPS is calculated.

Maximum connections

The maximum connections value is a hard limit. When concurrent connections exceed this limit, new connection attempts time out or fail.

The memory consumption per connection varies by workload, so actual connection capacity may differ from the listed maximum.

Recommended safe limit: Stay below LEAST({DBInstanceClassMemory/11MB}, 5000) to avoid memory pressure.

To check the current connection limit:

show max_connections;

To check the number of active connections:

select count(1) from pg_stat_activity;

If your workload requires more connections than your node supports, move to a node with more memory. Alternatively, use a connection pooler to multiplex application connections onto fewer database connections.

IOPS and cluster scaling

In an Enterprise Edition cluster, IOPS and I/O bandwidth scale proportionally with the number of nodes. Each node has its own independent storage specification — specifications are not pooled across nodes.

Example: A cluster with one read-write node and three read-only nodes using 8-core 32 GB nodes (PSL5) has a total cluster performance of 4 × 96,000 IOPS and 4 × 8 Gbps I/O bandwidth. Each individual node retains its own 128,000 IOPS and 8 Gbps I/O bandwidth.

How PSL4 IOPS is calculated

For PSL4 storage, effective node IOPS is determined by both the compute specification ceiling and your storage capacity:

Node IOPS    = min(maximum IOPS of the computing specification, storage IOPS)
Storage IOPS = min(1,800 + 50 × storage capacity in GB, 50,000)

Example: For a polar.o.x8.xlarge node (computing specification maximum IOPS: 80,000) with 100 GB of storage:

  1. Storage IOPS = min(1,800 + 50 × 100, 50,000) = min(6,800, 50,000) = 6,800

  2. Node IOPS = min(80,000, 6,800) = 6,800

To reach higher IOPS on PSL4, increase your storage capacity. The maximum storage IOPS for PSL4 is capped at 50,000 regardless of the node type.

Version note

In minor version 1.1.7 (released December 2020), the maximum number of connections changed for some compute node specifications. The table above shows the updated values. Clusters created after this release use the new values automatically. For existing clusters, change the cluster specifications to apply the updated connection limits.