Capacity / Cost
Derive event rates, partitions, retention, storage, compute costs, and map the stack to AWS plus open-source services.
On this page
Scale math
Make scale math, tool choices, and cost controls explicit and defensible.
I derive capacity from event math, then show how topic retention, compaction, cluster sizing, storage layout, and query engine choices control cost.
Quick jumps
Click to jumpMonthly users
200M-250M
Docs use both 200M+ and 250M assumptions
Daily active users
80M
Used for heartbeat and Kafka math
Peak concurrency
15M
Useful for QoE and live ops workloads
Derive capacity from event math, not guesses
Show the interviewer you can work from DAU and heartbeat frequency to partition count and storage.
Interactive scale calculator
Turn interview assumptions into throughput, storage, and partition math.
Heartbeats / user / day
240
watch_seconds_per_day / heartbeat_interval
Heartbeat events / day
19.20B
80M users x 240
Raw TB / day
34.6 TB
Only compressed events, before replication and long-term lifecycle
Avg events / sec
400.0K
daily_events / 86,400
Peak events / sec
2.00M
avg events/sec x peak multiplier
Kafka partitions
260
ceil(peak / per_partition x headroom)
Bronze hot storage
3.11 PB
Assumes 90 hot days of raw history
Silver retention
12.61 PB
Assumes Silver compresses to 50% and retains 2 years
Heartbeats per user/day
watch_seconds_per_day / heartbeat_interval_sec
7,200 / 30 = 240
Peak events/sec
daily_events / 86,400 x peak_multiplier
400.0K x 5 = 2.00M
Kafka partitions
ceil(peak_events_per_sec / safe_events_per_partition x headroom)
ceil(2.00M / 10000 x 1.3) = 260
Show where cost changes as the platform scales
Retention, partitions, cluster sizing, and serving engine choices all move spend.
Trade-off matrix
Kafka vs Kinesis
Option A
Option B
Recommendation
Why
When to change
Required trade-offs
Kafka = more flexibility
Kinesis = simpler managed path
Both need careful partition/shard math
Map the design to concrete services
Make the stack feel deployable, not abstract.
Event backbone
AWS
MSK
Open source
Kafka + Schema Registry
Why here
Streaming compute
AWS
Kinesis Data Analytics / EMR on EKS
Open source
Flink
Why here
Batch compute
AWS
EMR / Glue
Open source
Spark + dbt
Why here
Lakehouse format
AWS
S3 + Glue Catalog
Open source
Iceberg
Why here
Serving
AWS
Redshift / Athena
Open source
Trino / Pinot
Why here
Use a scale-and-cost line with real numbers
Good answers derive capacity instead of hand-waving it.
Say This In Interview
I derive capacity from event math, then show how topic retention, compaction, cluster sizing, storage layout, and query engine choices control cost.
30-second answer
FastI derive capacity from event math, then show how topic retention, compaction, cluster sizing, storage layout, and query engine choices control cost.
2-minute answer
BalancedDerive event rates, partitions, retention, storage, compute costs, and map the stack to AWS plus open-source services. Focus on PRODUCTION, 7 min, Make scale math, tool choices, and cost controls explicit and defensible..
Deep answer
DeepI derive capacity from event math, then show how topic retention, compaction, cluster sizing, storage layout, and query engine choices control cost. Then go layer by layer through PRODUCTION, 7 min, Make scale math, tool choices, and cost controls explicit and defensible., including trade-offs, failure handling, and why this shape is right for the consumer SLAs.
Last reviewed June 2026 · By Prasoon Parashar
Numbers are interview assumptions, not real Netflix internal figures.