Skip to content

Instantly share code, notes, and snippets.

@craigds
Last active February 26, 2026 01:34
Show Gist options
  • Select an option

  • Save craigds/8e5f6b576e4bcdc81d289a1265919a07 to your computer and use it in GitHub Desktop.

Select an option

Save craigds/8e5f6b576e4bcdc81d289a1265919a07 to your computer and use it in GitHub Desktop.

Total cost of datadb (monthly)

Component datadb-c datadb-d Combined
EC2 compute $1,064 $1,029 $2,094
EBS volumes $384 $384 $768
Total $1,448 $1,413 $2,862

Savings plans

The EC2 compute costs above are already post-savings-plan (effective) costs. The account has 4 active Compute Savings Plans (general-purpose, not locked to any instance family) with a combined commitment of ~$1,560/mo and ~100% utilization.

m6id coverage is ~91% — most of the datadb compute spend is covered by savings plans. Without them, the on-demand rate for m6id.4xlarge in ap-southeast-2 would be significantly higher. Since these are Compute Savings Plans, they'd automatically apply if the instances were changed to a different type.

LDS table disk usage

Sampled 20% of layer tables (5,344 of ~26,700) to estimate LDS vs non-LDS disk usage:

Category Tables (sampled) Size (sampled) Estimated total
LDS 1,173 397 GB ~992 GB
non-LDS 4,171 670 GB ~1,675 GB
Total 5,344 1,067 GB ~2,666 GB

LDS tables account for ~37% of datadb disk usage.

Potential savings from removing LDS tables

Current volume usage is ~2.7 TB per instance (91% of 3 TB). Removing ~1 TB of LDS data would reduce usage to ~1.7 TB, allowing volumes to be downsized from 3,000 GB to ~2,000 GB.

3,000 GB (current) 2,000 GB (without LDS)
Storage (gp3) $316.80 $211.20
IOPS (10k) $46.20 $46.20
Throughput (600 MB/s) $25.08 $25.08
Per volume $388.08 $282.48

Savings: ~$211/mo ($2,534/yr) across both instances — only ~7% of the total $2,862/mo datadb cost. Compute dominates at $2,094/mo.

Compute downsizing options

Removing LDS would significantly reduce query load, allowing downsizing from m6id.4xlarge. We don't need NVMe instance storage (data is on EBS), so switching to a non-d type is fine. For PostgreSQL, memory-optimized (r-series) is ideal since more RAM = more data cached in shared_buffers/OS page cache.

On-demand pricing in ap-southeast-2 (Compute Savings Plans apply automatically to all):

Instance vCPU RAM On-demand ~Monthly (2 instances)
m6id.4xlarge (current) 16 64 GB $1.193/hr $1,741
r6i.2xlarge 8 64 GB $0.604/hr $882
r7i.2xlarge 8 64 GB $0.634/hr $926
r6i.xlarge 4 32 GB $0.302/hr $441
m6i.2xlarge 8 32 GB $0.480/hr $701

Recommended: r6i.2xlarge — same 64 GB RAM as current, half the vCPUs (likely sufficient without LDS query load), ~49% cheaper on-demand. Savings Plans discount applies on top.

Estimated total savings from removing LDS

Component Current (2x m6id.4xlarge) Downsized (2x r6i.2xlarge) Savings
Compute $2,094 ~$1,047 ~$1,047
EBS volumes (3 TB → 2 TB) $768 ~$565 ~$203
Total $2,862 ~$1,612 ~$1,250/mo ($15,000/yr)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment