Amazon Timestream

Overview

Amazon Timestream is AWS's fully managed, serverless time-series database — optimized for storing and querying data where time is the primary axis, such as IoT sensor readings, application metrics, server telemetry, and financial tick data.

What is Time-Series Data?

Time-series data is a sequence of values recorded at successive points in time. Every record has three components:

[ timestamp ]  [ dimension(s) ]  [ measure(s) ]
  10:00:01       server="web-01"   cpu=72.4
  10:00:01       server="web-02"   cpu=45.1
  10:00:02       server="web-01"   cpu=73.1

Why not just use a regular relational database?

A general-purpose database can store time-series data, but time-series workloads have characteristics that make it inefficient:

Architecture

Two-Tier Automatic Storage

Timestream automatically manages a two-tier storage model — retention policies are configured and data moves between tiers automatically:

Write                     Memory Store                    Magnetic Store
  │       recent, hot data    │        older, cold data        │
  └──────▶  (in-memory)       │  ──auto-moves after TTL──▶  (S3-backed columnar)
            ms latency        │                               ms–seconds latency
            hours to days     │                               months to years

A memory store retention (e.g. 24 hours) and a magnetic store retention (e.g. 1 year) are configured per table. Data older than the memory retention threshold is automatically moved to magnetic storage. No manual partitioning or archival jobs needed.

Serverless Scaling

• No cluster to provision — Timestream automatically scales read and write throughput
• Billing is based on data written, data stored (per tier), and data queried
• No capacity planning required

Data Model

Timestream organizes data into databases → tables. Each table stores time-series records.

A record must have:

• A timestamp (nanosecond precision)
• One or more dimensions (string key-value pairs identifying the source)
• One or more measures (the numeric or string values being tracked)

Database: "infrastructure"
  Table: "server_metrics"
    Record:
      time:       2024-01-15 10:00:01.000000000
      dimensions: [server="web-01", region="us-east-1", az="us-east-1a"]
      measures:   [cpu_utilization=72.4, memory_used_gb=6.2, disk_io_ops=1240]

Dimensions are automatically indexed. Measures are stored as columnar data optimized for range queries over time.

Query Engine

Timestream uses a purpose-built SQL dialect with time-series specific functions that would be complex to write in standard SQL:

-- Average CPU per server over the last hour, in 5-minute buckets
SELECT server,
       bin(time, 5m) AS time_bucket,
       avg(cpu_utilization) AS avg_cpu
FROM "infrastructure"."server_metrics"
WHERE time BETWEEN ago(1h) AND now()
GROUP BY server, bin(time, 5m)
ORDER BY server, time_bucket;

Built-in time-series functions:

Common Use Cases

Timestream vs. Other AWS Databases

Integration with AWS Services

Cross-Cloud Equivalents

Pricing Model

• Writes: per million time-series data points written
• Memory store: per GB-hour stored
• Magnetic store: per GB-month stored (significantly cheaper than memory)
• Queries: per GB of data scanned
• No charge for the server/cluster — serverless

Related Services / See Also

• Amazon Kinesis and Managed Flink — stream IoT events into Timestream in real time
• Amazon DynamoDB and DocumentDB — NoSQL for non-time-series workloads
• Database Performance Fundamentals — OLTP/OLAP workload types
• Amazon Managed Grafana — visualization layer for Timestream metrics