Serverless sizing: Lambda, SQS, Kinesis, SNS and API Gateway

Lambda cost — requests a month, average duration, memory, architecture, the share of invocations that are cold starts and their INIT time (billed since 1 August 2025), /tmp size and provisioned concurrency. Shows the bill line by line, x86_64 against arm64 (with an optional speed difference), and the cost at every memory size for a workload that’s partly CPU-bound — CPU scales with memory, one vCPU at 1,769 MB, six at 10,240 MB.

Concurrency & cold starts — concurrency = requests per second × duration, checked against the account quota (1,000 by default), reserved concurrency and the 10 × concurrency requests-per-second limit, with the time Lambda takes to scale (1,000 more environments every 10 seconds per function) and how many requests a sudden burst sees throttled. Then what makes cold starts slow, and what helps.

SQS — messages, size (64 KB chunks, 1 MiB maximum), send batching, the Lambda batch size and window, function and visibility timeouts. Counts every API request including idle long polls, checks AWS’s rule of a visibility timeout of at least six times the function timeout plus the batch window, FIFO throughput (300 calls/s per action, 70,000 in high-throughput mode in the biggest regions) and message groups, and how fast the event source mapping ramps (300 a minute up to 1,250, or provisioned mode’s pollers).

Kinesis — records per second and size (or MB/s), peak factor, shared and enhanced fan-out consumers, retention and the Lambda parallelisation factor. Gives the shard count and what forced it (bytes, records or reads), and the monthly cost provisioned and on-demand.

SNS fan-out — publishes and subscriptions by protocol: SQS and Lambda deliveries are free, HTTP and email aren’t; filter policies, payload filtering and FIFO topics.

API Gateway — REST against HTTP APIs with the request tiers, the HTTP API’s 512 KB metering, REST caching, WebSockets, and a table of the features you give up for the lower price.

Prices: AWS price list files for AWS Lambda (published 2026-10-22), Amazon SQS (2026-09-11), Amazon Kinesis (2026-09-11), Amazon SNS (2026-09-15) and Amazon API Gateway (2026-10-13), read 2026-10-28.

Limits and rules: the Lambda Developer Guide (concurrency, scaling rate, requests-per-second limit; SQS event source mapping configuration and scaling), the Amazon SQS Developer Guide (message quotas, FIFO throughput) and the Kinesis Data Streams quotas page, read 2026-10-28; the AWS Compute Blog post on INIT billing.

The tests check the arithmetic against AWS’s own worked examples: the Lambda pricing page’s 512 MB × 3 million requests example ($18.73), the concurrency guide’s examples, and the Kinesis pricing page’s 100 records/s of 35 KB (4 shards, $43.80 of shard-hours).