Ready to Start?Discover our quick start guide and build your first application in 5-10 minutes
Quick Start Guide
Download IgniteARCHITECTURAL BACKBONE
CORE FEATURES
DATA PLACEMENT & PROCESSING
OPERATIONAL EXCELLENCE
ADDITIONAL USE CASES
Download IgniteARCHITECTURAL BACKBONE
CORE FEATURES
DATA PLACEMENT & PROCESSING
OPERATIONAL EXCELLENCE
ADDITIONAL USE CASES
Apache Ignite
Memory-First Storage Architecture
Data lives in memory by default. Disk provides durability.
Apache Ignite places data and operations in memory by default. Persistence provides durability rather than serving as the primary access path. This design delivers the latency characteristics needed for complex operations within shrinking transaction windows.
AIPERSIST: Persistent Storage with Memory-First Access
What It Provides
AIPERSIST keeps data in memory while maintaining durable copies on disk. The storage engine provides persistence with sub-millisecond latency for hot data. Data remains available after restarts without full reload.
How It Works
Data modifications replicate through the distributed consensus log (Raft). No traditional write-ahead log needed. The storage engine persists data in sorted structures that support efficient scans and range queries.
When to Use
Use AIPERSIST when you need durability without sacrificing latency. Suitable for transactional workloads, event stream processing, and operational data stores where restart tolerance matters.
Memory Management
The storage engine automatically manages memory allocation. Hot data stays in memory. Cold data resides on disk but loads to memory on access. No manual cache configuration needed.
AIMEM: Pure Memory Storage
What It Provides
AIMEM stores data entirely in memory with no disk persistence. This delivers microsecond-level access latency when durability is unnecessary. Data exists only while the cluster runs.
How It Works
Data replicates across cluster nodes for availability but never touches disk. Version chains support MVCC without persistence overhead. All coordination and transaction state stays in memory.
When to Use
Use AIMEM for session state, real-time aggregations, temporary result sets, or derived data that can be recreated. Appropriate when restart tolerance isn't required or data has external durability.
Availability
AIMEM provides high availability through replication. Data remains accessible during node failures as long as a majority of replicas are reachable. Losing the majority means data loss.
RocksDB: Experimental Storage
Current Status
RocksDB storage remains experimental. It provides an alternative persistence approach for specialized workloads. Not recommended for production deployments until the implementation matures.
Characteristics
RocksDB offers different performance trade-offs compared to AIPERSIST. It may suit workloads with specific access patterns or storage requirements. Evaluate carefully before adoption.
How Storage Connects to the Foundation
Raft-Based Durability
AIPERSIST uses Raft consensus for durability instead of traditional write-ahead logs. Data replicates through the distributed consensus log. This eliminates local WAL overhead while providing durability.
MVCC Version Chains
Both AIPERSIST and AIMEM leverage MVCC version chains. Multiple versions of each row enable snapshot isolation without blocking. Storage engines manage version cleanup automatically.
Memory-Resident Operations
SQL queries, transactions, and compute jobs execute directly against in-memory data. No cache-miss penalties. No cache-warming strategies. Data already resides where processing happens.
Ready to Start?
Read DocumentationLearn about table creation, storage engines, and configuration options
Tables Documentation