Skip to content
Cloud Efficiency Hub

Overprovisioned Base Capacity in Redshift Serverless Workgroups

The short version

Redshift Serverless bills compute in RPU-hours for the capacity in use while queries run, and the base capacity setting of a workgroup determines the capacity Redshift uses to serve queries.

PointFive Research

Cloud cost research at PointFive

AWS service
AWS Redshift
Category
Databases
Reference
CER-0366
Type
Overprovisioned Resource

Explanation

Why the waste happens and who it affects.

The default base capacity is 128 RPUs, and many workgroups are created with that default or with a value copied from a sizing exercise and never adjusted. When the workload is made of light dashboard queries, small ETL steps, zero-ETL refreshes or connection health checks, each active period is billed at the full base capacity even though a much smaller base would complete the same work at acceptable speed.

The effect is amplified by the billing rules: usage is metered per second with a 60-second minimum charge, and AWS notes that capacity can stay at a higher level for a period after load falls. The Redshift documentation states that a higher base capacity yields better performance but also incurs higher costs, and its zero-ETL cost guidance recommends the lower 8 RPU base capacity where available.

Billing model

The pricing dimensions that drive this cost.

Redshift Serverless compute is billed only while queries run, at a rate driven by the capacity in use.

RPU-hours
Compute billed for the capacity used in a given duration, in RPU-hours on a per-second basis, and not billed when no queries are running
Base capacity
The RPU level Redshift uses to serve queries, 128 RPUs by default and adjustable from 4 RPUs up to 512 (1024 in some Regions)
Minimum charge
Each period of usage is charged for at least 60 seconds of resource usage
Scaled capacity
Capacity added above the base for higher load is billed at the same RPU rate as base compute

How to detect

5 checks to find it in your estate.

  • List workgroups and their base capacity (get-workgroup or the Workgroup configuration Limits tab) and flag workgroups still at the 128 RPU default or at a high value with no documented reason
  • Query SYS_SERVERLESS_USAGE to see charged_seconds and compute_capacity per minute; workgroups whose usage consists of many short periods at full base capacity are paying the base rate for light work
  • Review ComputeCapacity and ComputeSeconds in CloudWatch (AWS/Redshift-Serverless) together with query runtimes in SYS_QUERY_HISTORY; consistently short queries on a large base suggest the base can be lowered
  • Check for connection pools or BI tools sending frequent health-check queries; Redshift Serverless treats them as billable user activity, so they trigger compute charges at the workgroup's capacity even when no real workload runs
  • Confirm managed storage size before planning a reduction, because it constrains the minimum base capacity

How to fix

5 ways to remove the waste.

  • Lower the base capacity with update-workgroup or the console, testing representative queries at each step; values are 4 RPUs, or multiples of 8 from 8 to 512, and editing the base capacity might cancel some running queries
  • Respect the documented limits: 4 RPUs supports up to 32 TB of managed storage, 8 or 16 RPUs up to 128 TB, and above 128 TB the base cannot be set below 32 RPUs; AWS also recommends 32 or more RPUs for tables with many columns and higher concurrency
  • For zero-ETL target workgroups, use the lower 8 RPU base capacity where available and lengthen REFRESH_INTERVAL where immediate freshness is not needed
  • Use the price-performance target (for example toward Optimizes for cost) for workloads in the 8 to 512 RPU range, and set Max capacity to cap scaling above the base
  • Set a Maximum RPU hours usage limit per day, week or month with a log, alert or turn-off-user-queries action, and reduce connection-pool health-check frequency to avoid paying for keep-alive queries

Documentation

Vendor references for pricing and configuration.