Future planning
Traditional database architectures rely on high-end hardware and a small number of servers. As business demands grow, these systems cannot meet the scaling requirements as new business develops. Cloud native databases address these constraints through distributed architectures and compute-storage separation, letting you scale resources to match your business requirements.
Next-generation database architecture
The next-generation enterprise database architecture combines three capabilities: a cloud native foundation, a distributed design, and hybrid transactional and analytical processing (HTAP).
Storage layer: distributed shared storage
Data is held in distributed shared storage, decoupled from compute. Compute nodes can scale up or down independently—adding capacity when demand spikes and releasing it when demand drops—without moving or resharding data. This separation means storage durability and compute performance are optimized independently.
Compute layer: shared-nothing architecture
A shared-nothing architecture partitions data across shards. In high-concurrency scenarios, the number of shards that need to be scanned is greatly decreased, which simplifies distributed transaction coordination.
Together, these two layers let you scale out on demand while maintaining high availability.
Trends shaping database architecture
Four trends are driving the adoption of this architecture:
Scalability and high availability
Business growth requires databases that scale without downtime or manual resharding. A distributed, cloud native architecture that combines distributed shared storage with compute-storage separation achieves this: it incorporates the features of both the cloud native model and the shared-nothing model, so you can scale out your databases based on your business requirements.
Heterogeneous data sources
Modern applications generate multi-model data: structured records, unstructured content, and high-dimensional vectors. Processing and integrating these types requires engines that handle data from heterogeneous sources and convert unstructured data into structured form using vector processing engines (VPEs).
Online real-time analytics
Analyzing large datasets has traditionally required offloading data to a separate system. Architectures that support DSP and massively parallel processing (MPP) enable interactive analysis directly on live data, with scheduling optimized for parallel execution.
Simplified operations and maintenance
As database clusters grow, manual operations and maintenance (O&M) becomes a bottleneck. A management and control platform providing intelligent scheduling, intelligent monitoring, and automated issue remediation reduces this burden. Built-in data encryption and privacy protection address security requirements.