Incremental cost / 7-day billing window
—
PNB-86763 · Amazon MSK
STOCK retention decision console
28-day topic-total traffic, regional AWS prices, 7D/8D retention, canned-reader egress, and Provisioned RF sensitivity in one auditable model.
5-minute cadenceUSD · 2026-07-29 list rates
Recommended: keep the existing MSK Serverless clusters
Selected Serverless monthly delta—
Monthly run-rate
—
8-day logical footprint
—
Projected monthly topic input
—
Measured 28-day input profile
STOCKSTOCK_EXTENDED
Architecture monthly comparison
Same scope, retention, readers, RF 3, and 20% storage headroom.
Selected calculation
Serverless base cluster and partition hours are excluded because they already exist and do not change with retention.
| Cost line | Formula | Monthly USD | Confidence |
|---|---|---|---|
| Selected monthly total / lower bound | |||
STOCK vs STOCK_EXTENDED
Measured NV topic-total expression series; these are different workloads and should not be substituted.
| Metric | STOCK | STOCK_EXTENDED | Interpretation |
|---|
The month export has no ap-southeast-1 STOCK or STOCK_EXTENDED series. SG STOCK is clearly marked as an estimate from the seven-day SG/NV ratio. SG STOCK_EXTENDED is not estimated because no SG source series was supplied.
How the model calculates cost
Retention target and replay requirement are separated so the decision is traceable.
Measured volume
Σ(BytesInPerSec × 300) ÷ 2³⁰Integrates every five-minute topic-total rate.Serverless storage delta
Daily GiB × (target days − 1) × $/GB-monthAWS manages Serverless replication; RF is not multiplied.New-reader egress
Monthly input GiB × readers × $/GB-outA full reader consumes every topic byte once.Provisioned primary storage
Daily GiB × days × RF × headroom × EBS rateRF applies only to Provisioned Standard physical capacity.Provisioned broker compute
3 brokers × 730 hours × regional m7g.large rateOne broker per AZ; a lower-bound dedicated cluster.7-day billing view
Monthly run-rate × 168 ÷ 730Prorates steady-state cost to seven days.Data quality and decision confidence
Measured facts, inferred aggregation, estimates, and missing inputs are kept separate.
High confidence · NV input8,064 complete five-minute points across exactly 28 days. All four month files are byte-identical; one canonical file is used.
High confidence · expression totalsOver the legacy overlap, e1 is ~2× STOCK Average and ~3× STOCK_EXTENDED Average. The e1 values are not multiplied again.
Medium confidence · SG STOCKSG month data is absent. The NV curve is scaled by the observed seven-day SG/NV ratio of .
Recommended Kafka topic config:
retention.ms=691200000 and retention.bytes=-1. Before change approval, export the CloudWatch widget source JSON, confirm current STOCK retention and partition count, and confirm whether both canned consumers replay the full topic. These checks refine the dollar amount; they do not make a new cluster economical for this retention change.Embedded source data
| Timestamp (UTC+6) | STOCK B/s | STOCK_EXTENDED B/s | STOCK GiB / 5m |
|---|