PROVIDER Β· Google wide-column NoSQL service
Google Cloud Bigtable
Cloud Bigtable is Google Cloud's fully managed wide-column service from the Bigtable paper, scaling single tables to petabytes with linear write throughput and HBase API compatibility for high-volume time-series and feature stores.
Pros
Cons
Product Information
Detailed Notes
Overview
Cloud Bigtable is Google Cloud's fully managed wide-column NoSQL service launched in 2015 out of the 2006 Bigtable paper, long powering Google search, ads, and analytics and exposing an HBase-compatible API as the external form of Google's internal sparse table.
For Backend & Database work its position is clear, since a single table reaches petabytes and billions of rows, write throughput scales linearly with nodes, and prefix reads are extremely fast, fitting time series metrics, IoT sensors, ad clicks, and recommendation features that scan by prefix under heavy write load.
The boundaries are query model and cost, because only single-row reads and prefix scans by row key are supported without secondary indexes or joins, billing follows node hours, clusters start at one node though production wants three for availability, and idle capacity never falls to zero.
Core Strengths
- Single tables reach petabytes and billions of rows with linear write scaling per node
- Row-key prefix scans are extremely fast for range reads and time-series aggregation
- An HBase-compatible API eases migration of existing HBase applications
- Fully managed with automatic multi-zone replication of nodes and storage
Product Ecosystem
Table model and row keys
Row keys define ordering and distribution and must follow query prefixes, with few column families and multi-version timestamped cells.
Clusters and performance
Throughput follows node count while SSD versus HDD storage changes latency and cost, and nodes scale online with load.
Ecosystem integration
Integration with Dataflow, Dataproc, and BigQuery supports streaming writes and offline analytics.
Drawbacks
- No secondary indexes or joins, only row-key access, making query changes expensive
- Node-hour billing never falls to zero under low load unlike serverless options
- Poor row key design creates hotspots whose fix requires rewriting data
- Tooling centres on Google Cloud, so cross-cloud migration rewrites the data layer
Use Cases
- Time-series metrics and IoT data (β β β β β ) - Prefix scans with high write throughput
- Ad and recommendation feature stores (β β β β ) - Low-latency wide-row reads
- Lifting HBase apps to cloud (β β β β ) - API compatibility eases migration
- Flexible multi-condition queries (β ) - No secondary indexes, need an index layer
Pricing
| Plan | Price | Scope |
|---|---|---|
| Single-node cluster | Per node-hour from about USD 0.65 | Development and testing only |
| Production cluster | Nodes times price plus storage | Three nodes recommended for availability |
| Storage types | SSD or HDD per GB-month | SSD for latency, HDD for savings |
Bigtable bills node-hours plus storage per gigabyte-month and nodes run even when idle, so cost is predictable but inelastic: single-node HDD suits development while production node counts should follow measured throughput to avoid over-provisioning.
FAQ
-
Bigtable or BigQuery? Bigtable for high-frequency point reads, prefix scans, and write-heavy ingestion, BigQuery for SQL analytics, joins, and ad hoc queries, often combined.
-
How should row keys be designed? Lead with the most common query prefix and avoid timestamps or sequential IDs at the start, using hashed prefixes to spread writes.
-
Can it replace HBase? The HBase API supports most read and write logic, though applications relying on coprocessors or custom filters need rework.