Built on a proprietary distributed storage engine, Tablestore ensures consistent low latency and high concurrent throughput, making it ideal for both latency-sensitive online services and high-throughput big data analytics.
Low-latency read and write performance
Millisecond response: For massive datasets, the average read and write latency (P99) for a single row is under 10 ms. This supports the real-time access needs of online applications.
High concurrent throughput: The distributed architecture supports horizontal scaling to handle high-concurrency read and write requests of up to tens of millions of queries per second (QPS).
Data retrieval performance
Tablestore delivers high performance across various retrieval scenarios:
Large-scale retrieval: Provides millisecond-level retrieval responses for datasets with tens of billions of records.
Comparison with HBase: For point queries and batch queries, Tablestore performs 3 to 9 times faster than open source HBase.
Comparison with Elasticsearch: Using inverted indexes, Tablestore reduces data retrieval latency by more than 25% compared to open source Elasticsearch.
Vector search performance
Tablestore provides specialized optimizations for vector search in AI scenarios:
Index building: Tablestore supports the real-time, stream-based building of vector indexes. This method is 65% more efficient and more cost-effective than using open source vector engines.
Query response: In high-dimensional vector search scenarios, query latency is one-sixth that of open source vector engines. This improves the response speed of AI applications such as retrieval-augmented generation (RAG) and recommendation systems.