An Overview of Time Series Databases and Tencent CTSDB

With the rapid growth of the Internet, big data, and IoT, many data streams are now time‑related, such as step counts, stock prices, and bike‑sharing device statuses. To store these time‑series data, enterprises have launched dedicated time‑series databases.

What is a Time‑Series Database?

Time‑series data are observations recorded in chronological order, commonly found in IT infrastructure, monitoring systems, and IoT. By linking isolated observations along a timeline, trends and anomalies become visible, enabling reporting, statistical analysis, and forecasting.

Time‑Series Data Model

metric – the dataset (similar to a table)

point – a single record (similar to a row)

timestamp – the time of collection

tag – dimension columns that identify the source (e.g., host, port)

field – measured values that vary over time

Typical data include a metric name (e.g., Network), a timestamp, two tags (host, port), and two fields (bytes_in, bytes_out). Each host/port generates a point every half hour, and the fields change smoothly over time.

Characteristics of Time‑Series Data

Data pattern: continuously growing, repeated dimensions, smooth metric changes.

Write workload: massive high‑concurrency writes with little or no updates.

Query workload: frequent recent‑data queries, aggregation‑heavy analysis.

Traditional Solutions

Relational databases (e.g., MySQL) suffer from high storage cost, maintenance overhead, low write throughput, and poor analytical performance for massive time‑series workloads. Hadoop‑based stacks (Hadoop, Spark) have high latency and slow query performance due to batch processing.

Time‑Series Databases

Specialized time‑series databases optimize storage, write, and query paths for the unique properties of time‑series data, achieving lower storage cost, high‑concurrency writes, and low‑latency queries.

CTSDB Overview

Tencent Cloud Time Series Database (CTSDB) is a distributed, high‑performance, sharded, self‑balancing time‑series database. Its key features include:

High performance: batch writes, high‑concurrency queries, linear scalability.

High reliability: multi‑replica, rack‑aware placement.

Ease of use: JSON‑based REST API, rich data types.

Low cost: columnar storage with ~0.1 compression ratio, data down‑sampling.

Powerful aggregation: max, min, avg, percentile, sum, count, group‑by, script aggregation, time‑range, geo, nested aggregation.

Additional capabilities: monitoring & alerting, permission system, data expiration, export.

Performance Comparison

CTSDB was benchmarked against InfluxDB using the official InfluxDB comparison tool. Tests covered write throughput (single‑node CTSDB vs. InfluxDB, single‑node vs. dual‑node CTSDB) and query latency under varying concurrency.

Write Performance

CTSDB single‑node achieved up to 190,000 writes per second, surpassing InfluxDB’s 150,000. Dual‑node CTSDB reached 340,000 writes per second, demonstrating linear scalability.

Query Performance

CTSDB’s query QPS consistently outperformed InfluxDB; at high concurrency (40 threads) CTSDB was nearly four times faster (~20,000 QPS) while InfluxDB failed at 50 threads. Dual‑node CTSDB further improved query throughput, showing near‑linear scaling.

Current Status and Future

CTSDB powers more than 20 core Tencent services, handling tens of thousands of writes per second and over 20 TB of daily data. It will soon be launched on Tencent Cloud to serve the IoT industry, with ongoing improvements in cost, usability, and functionality.