What is AnalyticDB
AnalyticDB is a cloud-native real-time data warehouse from Alibaba Cloud. It enables real-time data ingestion from sources such as Online Transactional Processing (OLTP) databases and log files, and it can perform petabyte-scale data analysis in seconds. AnalyticDB features a cloud-native storage-compute decoupled architecture. This architecture supports pay-as-you-go storage, elastic scaling for compute resources, and resource isolation for offline processing and online analytics. This design helps enterprises improve data processing efficiency, control costs, and ensure system stability. AnalyticDB is compatible with the MySQL, PostgreSQL, and Spark ecosystems.
AnalyticDB has two engines: AnalyticDB for MySQL and AnalyticDB for PostgreSQL.
|
Comparison Criterion |
AnalyticDB for MySQL |
AnalyticDB for PostgreSQL |
|
|
Ecosystem |
Highly compatible with MySQL Highly compatible with Spark |
100% compatible with PostgreSQL Highly compatible with Oracle |
|
|
Edition |
|
|
|
|
Architecture |
Storage-compute decoupled architecture |
||
|
Extensibility |
Common features |
Vertical scaling Horizontal scaling |
|
|
Differences |
Supports the Multi-Cluster model for automatic resource scaling Supports the MIN-MAX model for time-based and automatic elastic scaling |
Supports scheduled tasks for scheduled automatic scaling Supports a serverless model for on-demand automatic scaling |
|
|
Key features |
Common features |
AISearch Full-text index Offline batch processing Real-time materialized view |
|
|
Differences |
Data lake Spark offline processing Intelligent diagnosis and tuning for query performance |
Retrieval-Augmented Generation (RAG) applications Spatiotemporal analysis |
|
|
Scenarios |
Common features |
Real-time data warehouse Real-time log analysis Business intelligence (BI) reports |
|
|
Differences |
Precision marketing Federated analytics across multiple sources Big data storage and analytics Offline data acceleration Migration from other data lake or data warehouse services (such as Databricks, Athena, self-managed Spark, or Presto) |
One-stop solution for building Large Language Model (LLM) applications Exclusive enterprise knowledge bases Geographic Information System (GIS) spatiotemporal big data analytics Integrated offline and online data analytics Migration from other cloud data warehouse services (such as Greenplum, Redshift, Synapse, Snowflake, or BigQuery) |
|
|
Customer industries |
Gaming, retail, automotive |
Retail, e-commerce, education |
|
|
Cost savings |
Common features |
Data storage fees are based on the actual data volume. Tiered storage for hot and cold data reduces storage costs. Automatically scales resources up or down based on predictable business peaks and troughs. This ensures sufficient resources during peak traffic and prevents resource waste after peaks. |
|
|
Differences |
Automatically scales resources up or down based on the actual business workload. |
Manually start or pause instances as needed. |
|
Introduction to AnalyticDB for MySQL
Data sources
The APS data pipeline provides a one-click and low-cost method to ingest data from databases, logs, and big data systems.
Storage layer + Compute layer
It supports its self-developed Xihe compute engine and Xuanwu storage engine. It also integrates the open source Spark compute engine and Hudi storage engine. These open source engines support a wider range of data analytics scenarios. Mutual access between the self-developed and open source engines is enabled, providing a more integrated experience.
-
Storage layer: A single copy of the full data can meet the needs of both offline and online scenarios.
Online analytics scenarios require high-performance storage media to improve performance, while offline scenarios require low-cost storage media to reduce storage costs. To meet these different needs, a full copy of the data is first stored on low-cost, high-throughput storage media. This allows low-cost offline processing scenarios to directly read from and write to this data, which reduces data storage and data I/O costs and ensures high throughput. Then, real-time data is stored on separate storage I/O nodes (EIU) to ensure row-level real-time performance. Indexes are built on the full data and caching is used to accelerate data access. This design meets the needs of high-performance online analytics scenarios, delivering response times in the hundreds of milliseconds.
-
Compute layer: The Xihe compute engine intelligently selects the computation mode, and the open source Spark compute engine meets the needs of various scenarios.
The Xihe compute engine provides both Massively Parallel Processing (MPP) and Bulk Synchronous Parallel (BSP) modes. The MPP mode is a stream processing model and is not suitable for low-cost, high-throughput offline processing scenarios. The BSP mode uses a Directed Acyclic Graph (DAG) to split tasks and schedule them in batches. This approach supports large-scale data computation with limited resources and allows computation data to be written to disk. The Xihe compute engine includes an automatic switchover feature. If a query in MPP mode does not complete within a specified time, the system automatically switches to BSP mode for execution.
The open source Spark compute engine can meet the needs of more complex offline processing and machine learning scenarios. The Spark compute layer and storage layer are interconnected, allowing compute layer resources to be used to process storage layer data. This simplifies the creation and configuration of Spark resource groups.
Access layer
The access layer improves development efficiency by providing a unified billing unit, unified metadata and permissions, a unified development language, and a unified transmission link.
For more information about the AnalyticDB for MySQL product editions (Data Lakehouse Edition, and Data Warehouse Edition), see Product editions.
Introduction to AnalyticDB for PostgreSQL
AnalyticDB for PostgreSQL has two product forms: storage-elastic mode and Serverless. The storage-elastic mode uses a Shared-Nothing architecture based on ECS and ESSD cloud disks with MPP. Serverless uses a storage-compute decoupled Shared-Storage architecture based on ECS, local cache, and OSS remote storage.
An AnalyticDB for PostgreSQL instance consists of a coordinator node (also known as a master node) and multiple worker nodes (also known as segment nodes). The coordinator node manages metadata and load balancing. Worker nodes handle data processing and include the Orca optimizer, the proprietary Laser execution engine, and the Beam storage engine for high query performance. Combined with the IMV real-time materialized view component, AnalyticDB for PostgreSQL builds a real-time data warehouse. ESSD cloud disks attached to worker nodes store hot data, while cold data is stored in OSS. This hot-cold data tiering balances query performance and storage cost. AnalyticDB for PostgreSQL uses a compute-storage decoupled architecture. Compute resources on worker nodes support flexible scaling to optimize performance and resource utilization. However, storage resources only support scale-out and do not support scale-in. This design ensures data durability and system stability while allowing flexible storage capacity growth as data volume increases.