PNB-86284 · NV / SG cost analysis

STOCK 7-day replay cost explorer

Measured 28-day NV traffic plus the FeedOps-confirmed STOCK configuration. Compare the current byte-capped setup, the recommended 8-day policy, MSK Provisioned, tiered storage and Kafka on EC2.

5-minute CloudWatch dataSG values estimated

AWS console + FeedOps-confirmed NV facts

● Active Serverless

TopicSTOCK

Partitions2

Replication / min ISRRF3 · min ISR2

Configured time14 days

Size cap250 GiB / partition

Replica healthAll replicas in ISR

Important: 14 days is the configured time limit, not guaranteed history. The 250 GiB per-partition cap can delete older segments first. At the measured average and a balanced split, the cap is reached at about —.

✓

Recommendation

Selected monthly cost—

Scenario controls

Change every unavailable value here

Scope Topic Cost view Architecture Current retention timeConfirmed NV: 14D Target retention (days) Feed Replay replicas QS connections / replica

Advanced assumptions · partitions, RF, network and EC2

Actual STOCK partitions Confirmed NV: 2 Current cap / partition (GiB)Confirmed NV: 250; -1 = unlimited Target cap / partition (GiB)Recommended: -1 = unlimited SG / NV traffic ratio (%) From earlier 7-day overlap Replication factor Provisioned / EC2 only Storage headroom Broker / EC2 nodes Client network $ / GiB Default 0: same-region private path EC2 cross-AZ $ / GiB EC2 price multiplier EC2 operations $ / month Staff/support burden; default 0

Retention policy decision

Current vs cost-effective target

NV uses measured STOCK input and two confirmed partitions. Values are logical Serverless storage.

Policy	Time	Byte cap / partition	Estimated retained	Effective history	Storage / month	Decision

7-day billing view—

Monthly run-rate—

Target footprint—

Projected monthly input—

Observed monthly output—Included only in full workload view

Average message rate—

Measured NV profile

Traffic over 28 days

STOCKSTOCK_EXTENDED

Comparable scenario

Monthly architecture cost

Uses the same scope, target days, reader count and advanced assumptions.

Auditable result

Selected cost breakdown

Cost line	Formula	Monthly USD	Confidence
Selected total

Workload comparison

STOCK vs STOCK_EXTENDED

Metric	STOCK	STOCK_EXTENDED

Model boundaries

Confirmed, derived and assumed

ConfirmedNV STOCK has 2 partitions, RF3, min ISR2, healthy ISR, retention.ms=1209600000 (14D), retention.bytes=268435456000 (250 GiB/partition), cleanup.policy=delete.

Measured / derived28-day NV STOCK input/output/messages; effective retention and costs are derived from the observed aggregate input profile.

Editable / still conditionalOne full Feed Replay reader, SG/NV ratio, SG topic configuration, network $0, Provisioned sizing and EC2 operations. Partition skew and compression are not present in the exports.

Calculation rules

Exactly what is counted

Measured GiBΣ(rate × 300 seconds) ÷ 2³⁰Uses topic-total expression values.

Serverless incrementaltarget retained storage − current retained storage + new reader output + networkThe storage delta may be negative when replacing the current 500 GiB aggregate cap with an 8-day policy.

Serverless full viewcluster + partitions + input + observed output + new output + target storageShows every available MSK in/out line.

New reader outputmonthly input × replicas × QS connectionsFeed Replay creates consumers per connection; change both controls to match deployment.

Provisioned storagedaily input × target days × RF × headroomServerless storage is not RF-multiplied.

EC2 lower boundnodes + EBS + cross-AZ replication + network + operationsNo managed Kafka fee; operational risk remains.

Recommended change: reuse the existing Serverless STOCK topic and set retention.ms=691200000 plus retention.bytes=-1. Exactly 7 days is 604800000, but it puts the consumer on the deletion boundary. Validate oldest available timestamps and per-partition skew after the change.

Embedded evidence

Raw selected topic values

Timestamp	Bytes in / sec	Bytes out / sec	Messages / sec	GiB in / 5m