Tablestore leverages a distributed architecture and optimized storage algorithms to ensure high performance and reliability. This efficiency also enables fine-grained control over storage and compute costs, significantly reducing your overall IT expenses.
Storage cost control
Multiple replicas, single-replica billing: Tablestore uses a multi-replica distributed architecture to ensure data reliability. You are billed for the storage capacity of only a single data replica, which helps control costs.
Data compression: For time series models, Tablestore provides a data compression feature that can save up to 95% of your storage space.
Compute cost optimization
VCU mode: The new VCU mode lets you combine reserved VCUs and elastic pay-as-you-go VCUs to reduce computing costs. For testing purposes, configure 0 reserved VCUs.
Vector search cost benefits
Memory usage: For vector search, Tablestore uses the DiskANN algorithm. With this algorithm, only 10% of the graph index data is loaded into memory, and the remaining 90% is stored on disk. This method achieves a recall rate and performance similar to HNSW while reducing memory costs to only 10% of what HNSW requires.