About Aurora PostgreSQL I/O Pricing

Aurora costs are broadly classified into the following 4 categories. Other RDBMS database services like RDS PostgreSQL do not incur I/O charges, which may be why Aurora I/O charges are easy to overlook in cost estimates. (I have a sense that I/O charges range from a few percent to at most 10% of instance charges, but depending on how the database is used, it could be higher.)

1. Instance charges
2. Database storage
3. I/O charges
4. Data transfer

For details, please refer to the following. Usage charges for cloning and global databases are omitted as they depend on usage patterns.

Pricing - Amazon Aurora | AWS https://aws.amazon.com/jp/rds/aurora/pricing/

For the Tokyo region, I/O charges are 0.24 USD per 1 million requests. Also, 1 I/O operation is charged for up to 4KB of changes per data page. If a workload generates 2,000 IOPS per second, the monthly cost is calculated as follows. (Note: 2,000 IOPS per second is just a rough example.)

1,000 Reads/Second + 1,000 Writes/Second = 2,000 Number of I/Os per second
2,000 I/Os per second x 730 hours x 60 minutes x 60 Seconds = 5,256,000,000 Number of I/Os per month
5,256,000,000 x 0.00000024 USD = 1,261.44 USD (I/O Rate Cost)

Pricing itself can be calculated using the Pricing Calculator.

• AWS Pricing Calculator
  • https://calculator.aws/#/createCalculator/AuroraPostgreSQL

When Aurora I/O charges are high, check the following:

• Check Billed IOPS in CloudWatch
  • When there are spikes, check what operations are happening and whether I/O volume can be reduced
    • pg_dump, batch processing, full scans, etc.
• Use Performance Insights to identify SQLs generating IO: DataFileRead or IO:DataFilePrefetch events
• Check the buffer cache hit rate
  • Since data file I/O is occurring, adjust parameters to maximize in-memory processing

Actions to reduce I/O charges primarily include:

• Increase memory to enable in-memory processing
• Tune full scan SQLs
• Use indexes and partitioning to reduce unnecessary I/O

Update (2022/02/02)

The following article is also helpful - in fact, it’s better to carefully read the AWS Database Blog officially published in January 2022.

• Planning I/O in Amazon Aurora | AWS Database Blog