DEV Community

Sidharth Bangera
Sidharth Bangera

Posted on

Simulating DynamoDB Costs in Tables: GSIs, Backups, Streams, and Replicas

#dynamodb #aws #finops #cloud

Introduction

When I evaluate a DynamoDB design change, I want to understand how it affects cost before I change the AWS resources themselves.

In this walkthrough, I use Tables by ServerlessCreed, a DynamoDB GUI with cost analysis and simulation tools, to compare a baseline workload with several what-if changes: two global secondary indexes (GSIs), point-in-time recovery (PITR), backup storage, DynamoDB Streams reads, and two replica Regions.

The test uses a fixed synthetic workload so that each cost driver can be added separately and compared with the previous state. The dollar values shown here are simulator estimates, not actual AWS charges.

Prerequisites

To reproduce the workflow, you need:

  1. 1. Tables by ServerlessCreed installed. The installer is available from the official download page.
  2. Access to a non-production DynamoDB environment.
  3. A table available in Tables for cost analysis.
  4. A known Region and a consistent simulation period.

For this test, I used:

  • Tables version: v3.3.37
  • Table: Export_Source_Test
  • Primary Region: us-east-1
  • Simulation period: 7 days
  • Table items: 100
  • Execution: local what-if simulation

The simulation changed only estimator inputs; it did not modify those AWS resources.

1. Starting from the observed cost baseline

I first opened Cost Estimation using the table’s observed metrics.

Reads, writes, and table storage were available, but Streams reads and backup/PITR inputs were incomplete. Tables marked those components as incomplete instead of silently treating them as zero.

For the 7-day observed view, the current and forecast cost remained below $0.01. I used this only as the observed starting point; the detailed comparisons below use a fixed synthetic baseline.

Tables Cost Optimizer for Export_Source_Test in us-east-1, showing a 7-day observed baseline with current spend below $0.01, forecast below $0.01, write amplification 1.0×, and no material savings found.Observed cost baseline for Export_Source_Test. The 7-day Cost Optimizer view shows current and forecast spend below $0.01, write amplification of 1.0×, and no material savings identified before any simulation inputs were added.

2. Loading a fixed synthetic baseline

To make the comparison repeatable, I loaded the built-in synthetic baseline.

Input Value
Table storage 10 GB
Table read units 6,048,000 over 7 days
Table write units 3,024,000 over 7 days
Peak reads 20/second
Peak writes 10/second
Stream GetRecords calls 0
Backup storage 0 GB
PITR Off
Simulated GSIs None
Replica Regions None

With those inputs, Tables produced these monthly estimates:

Billing model Estimated monthly cost
On-Demand $13.84
Provisioned $10.47
Reserved 1 year $6.16
Reserved 3 year $4.33

For the comparisons below, I used On-Demand consistently so that each change was measured against the same billing model.

Tables Cost Estimation synthetic baseline showing 10 GB storage, 6,048,000 read units, 3,024,000 write units, 20 peak reads per second, 10 peak writes per second, zero Stream GetRecords calls, zero backup storage, PITR off, and no GSIs or replicas.

Tables Cost Estimation baseline pricing showing On-Demand $13.84, Provisioned $10.47, Reserved 1 year $6.16, and Reserved 3 year $4.33 per month.Synthetic baseline inputs. I loaded a fixed 7-day workload with 10 GB of storage, 6,048,000 read units, 3,024,000 write units, no Streams activity, no backup storage, PITR disabled, and no simulated GSIs or replicas.

3. Comparing two simulated GSIs

I next added two simulated GSIs with the same write volume but very different read activity.

For SimulatedGSI1, I used:

Read units: 500,000 over 7 days
Write units: 3,024,000 over 7 days
Enter fullscreen mode Exit fullscreen mode

The On-Demand estimate increased from $13.84 to $22.21/month, an increment of:

+$8.37/month
Enter fullscreen mode Exit fullscreen mode

The breakdown showed approximately $0.27/month for GSI reads and $8.10/month for GSI writes.

For SimulatedGSI2, I used:

Read units: 5,000 over 7 days
Write units: 3,024,000 over 7 days
Enter fullscreen mode Exit fullscreen mode

With both GSIs present, the estimate reached $30.31/month. The second GSI added approximately:

+$8.10/month
Enter fullscreen mode Exit fullscreen mode

Tables GSI simulation showing two GSIs with different read volumes and the same write volume.

Tables GSI simulation showing two GSIs with different read volumes and the same write volume.GSI simulation inputs. SimulatedGSI1 uses 500,000 read units and SimulatedGSI2 uses 5,000 read units over the 7-day period, while both use 3,024,000 write units.

Tables cost breakdown showing both simulated GSIs and an On-Demand estimate of $30.31 per month, with GSI write costs dominating the index cost.

Tables cost breakdown showing both simulated GSIs and an On-Demand estimate of $30.31 per month, with GSI write costs dominating the index cost.GSI cost result. With both simulated indexes added, the On-Demand estimate reaches $30.31/month. The similar write volume keeps the maintenance cost of the two GSIs close despite the large difference in read usage.

This produced the clearest ROI comparison in the test.

SimulatedGSI2 served 100 times fewer simulated reads than SimulatedGSI1, but it carried the same write volume and almost the same incremental cost. For this workload, that makes the second GSI the weaker-ROI candidate.

That does not mean a low-read production index should automatically be removed. It may still support an infrequent but important access pattern. The comparison instead gives a concrete reason to review whether its usage justifies the ongoing cost.

4. Adding PITR, backup storage, and Streams reads

I then added three more cost drivers one at a time.

Change Estimate before Estimate after Increment
Enable PITR $30.31 $32.31 +$2.00/month
Add 20 GB backup storage $32.31 $34.31 +$2.00/month
Add 1,000,000 Stream GetRecords calls over 7 days $34.31 $35.17 +$0.86/month

PITR enabled in the simulation. Point-in-time recovery was enabled while the existing GSI assumptions remained unchanged.

Tables cost breakdown showing PITR enabled, a Backups plus PITR cost of $2.00, and an On-Demand monthly total of $32.31.PITR cost result. Enabling PITR increased the simulated On-Demand estimate from $30.31 to $32.31/month, an increment of $2.00/month.

I then kept PITR enabled and added 20 GB of backup storage.

Backup storage added. Backup storage was increased to 20 GB while PITR and the existing GSI assumptions remained unchanged.

Tables cost breakdown showing 20 GB of backup storage with PITR enabled, a combined Backups plus PITR cost of $4.00, and an On-Demand total of $34.31 per month.Backup storage cost result. Adding 20 GB of backup storage increased the simulated On-Demand estimate from $32.31 to $34.31/month, another $2.00/month increase.

Finally, I added 1,000,000 Stream GetRecords calls over the same seven-day period. This increased the simulated On-Demand estimate from $34.31 to $35.17/month, an additional $0.86/month.

Tables Cost Estimation showing one million Stream GetRecords calls over seven days, with two simulated GSIs, PITR enabled, and 20 GB of backup storage.Streams read activity added. I set the simulation to 1,000,000 DynamoDB Streams GetRecords calls over the 7-day period while keeping the existing GSIs, PITR, and backup assumptions unchanged.

Tables cost breakdown showing one million Stream GetRecords calls and an On-Demand monthly estimate of $35.17 after PITR and 20 GB of backup storage were already included.Streams cost result. Adding 1,000,000 Stream GetRecords calls increased the simulated On-Demand estimate from $34.31 to $35.17/month, an increment of $0.86/month.

AWS documents PITR and on-demand backup charges as storage-based. DynamoDB Streams reads are billed in Streams read request units, with each GetRecords call counting as one unit, subject to the applicable AWS pricing and Free Tier conditions.

The $0.86 value above is the Tables simulator estimate for this seven-day Streams input.

Before adding replicas, the simulated On-Demand total was:

$35.17/month
Enter fullscreen mode Exit fullscreen mode

5. Adding two replica Regions

I then added two replica Regions:

Replica 1: us-east-2
Replica 2: us-west-1
Enter fullscreen mode Exit fullscreen mode

After adding us-east-2, the On-Demand estimate increased from $35.17 to $61.97/month.

Increment: +$26.80/month
Enter fullscreen mode Exit fullscreen mode

The breakdown showed $24.30/month for replicated writes and $2.50/month for replicated storage.

Tables Cost Estimation showing us-east-2 added as the first replica Region with two simulated GSIs, PITR enabled, 20 GB of backup storage, and one million Stream GetRecords calls.First replica Region added. I added us-east-2 as the first simulated replica while keeping the existing GSIs, PITR, backup storage, and Streams assumptions unchanged.

Tables cost breakdown after adding the us-east-2 replica, showing an On-Demand estimate of $61.97 per month and replica-related write and storage costs.First replica cost result. Adding us-east-2 increased the simulated On-Demand estimate from $35.17 to $61.97/month, an increment of $26.80/month.

After adding us-west-1, the estimate increased again:

$61.97/month -> $91.79/month
Enter fullscreen mode Exit fullscreen mode

The second replica added:

+$29.82/month
Enter fullscreen mode Exit fullscreen mode

With both replicas present, Tables showed $51.32/month for replicated writes and $5.30/month for replicated storage. The combined replica increase was therefore:

+$56.62/month
Enter fullscreen mode Exit fullscreen mode

Tables Cost Estimation showing two replica Regions, us-east-2 and us-west-1, with the existing GSIs, PITR, backup storage, and Streams assumptions still configured.Second replica Region added. I then added us-west-1 as a second simulated replica while keeping the rest of the workload unchanged.

Tables cost breakdown with replica Regions us-east-2 and us-west-1, showing a final On-Demand estimate of $91.79 per month with replicated write and storage costs.Two-replica cost result. Adding us-west-1 increased the simulated On-Demand estimate from $61.97 to $91.79/month. The two replicas added $56.62/month in total compared with the pre-replica state.

The replica increments differ because the selected Regions use different regional pricing.

This behavior is consistent with AWS DynamoDB Global Tables billing, where replicated writes are charged in every Region containing a replica and each replica also carries storage cost.

Tables also showed the one-year and three-year reserved options as Not offered after the Global Table scenario was added. AWS documentation states that reserved capacity cannot be purchased for replicated write capacity units.

Practical takeaways

The synthetic baseline started at $13.84/month On-Demand. After adding two GSIs, PITR, backup storage, Streams reads, and two replica Regions, the final simulated total reached $91.79/month.

The component-level comparison was more useful than the final number alone.

For this workload:

  • GSI write activity contributed much more to index cost than the difference in read activity.
  • The lower-read GSI had nearly the same maintenance cost as the higher-read GSI because both used the same simulated write volume.
  • PITR, backup storage, and Streams reads produced smaller incremental increases.
  • The two replica Regions were the largest incremental cost driver in the simulation.

These results are specific to the assumptions used in this test. Production traffic, item sizes, table class, Free Tier eligibility, discounts, Region-specific pricing, and the business value of individual access patterns can all change the outcome.

For me, the useful part of this workflow is being able to compare these design choices before changing the infrastructure. The simulation does not replace real usage data or an AWS bill, but it provides a consistent way to test assumptions and identify which cost drivers deserve closer review.

Top comments (0)