On July 16, 2024, TDBC Conference organized by China Academy of Information and Communications Technology(CAICT) was grandly held in Beijing. Themed "Independence, Innovation, Leadership" the conference gathered nearly a hundred experts and scholars in the field of databases, bringing high-quality insights into database technology and practical experience.
During the Conference, the CAICT officially announced the results of the "Trusted Database" evaluation for the first half of 2024. OpenPie's cloud-native vector database PieCloudVector passed the capability evaluation for vector databases and obtained certificate. In addition, OpenPie was selected for the "China Database Industry Atlas (2024)."
So far, among all the products that participated in the vector database evaluation of the CAICT, only two have passed all the test items, and PieCloudVector is one of them.
The "Trusted Big Data" series of evaluation tests is an authoritative testing system for big data products, aimed at comprehensively measuring the capabilities of enterprise-level big data products from the dimensions of basic capabilities, performance, reliability, security, etc.
This evaluation was based on the "Technical Requirements for Vector Databases," and after strict testing, PieCloudVector performed excellently in seven major capabilities: basic functions, operation and maintenance, security, compatibility, scalability, high availability, and tool ecosystem. The test results show that PieCloudVector meets the standard requirements in terms of the completeness, usability, and generality of vector database functions.
According to the CAICT: "The historical evaluation results of all tested products show that 'backup and recovery,' 'data lifecycle,' 'computing heterogeneous vector indexing,' 'data encryption,' and 'multi-modal data vectorization capabilities' are the test items with the lowest pass rates. The overall pass rate of all test items is 90.07%, and the average pass rate of optional items is 76.67%." PieCloudVector passed all test items with outstanding results in this evaluation, fully proving the capabilities of PieCloudVector.
PieCloudVector, as the second cloud-native vector computing engine of OpenPie's large model data computing system (PieDataCS), is an upgrade of the analytical database in the era of large models.
Main Features of PieCloudVector
PieCloudVector supports mainstream vector indexing, including IVF, HNSW, hybrid indexing, and Binary indexing, and has optimized the creation of vector indexing with multi-threading. The single-node multi-threading mode can fully schedule all computing resources, greatly improving the efficiency of index creation. It supports L2 distance, inner product, cosine similarity, Jaccard, and Hamming, with excellent key performance indicators (QPS, recall rate, and latency) for retrieval.
PieCloudVector can not only handle vector data but also scalar data, supporting mainstream vector retrieval KNN-ANN algorithms.
Highly compatible with SQL:2016 standard, fully supports SQL:1992 standard, and most of SQL:1999 standard and some SQL:2003 standard (mainly supporting OLAP features); compatible with PostgreSQL protocol, supporting standard database interfaces (ODBC, JDBC, etc.).
Supports mainstream hardware such as CPU and GPU with related performance optimizations.
Supports built-in Embedding functions, which can be extended and integrated with large language models (LLMs) as needed, supporting various methods such as matrix factorization, content-based embedding, item sequence-based embedding, and graph-based embedding.
Based on vector compression (Product Quantization, PQ) technology, it stores and compresses various types of original vector data (including images, videos, audio, text, and matrices) to reduce storage space, achieving more effective memory management while processing large datasets, and accelerating the speed of similarity search and nearest neighbor search.
Humanized operation and maintenance management interface, not only providing monitoring and alert functions, but also having a complete set of cluster and host monitoring indicators, while supporting query monitoring, log collection and analysis, and database operations, achieving intelligent operation and maintenance management.
Core Advantages of PieCloudVector
Main Application Scenarios of PieCloudVector