The rise of unstructured data retrieval in the AI era has created an urgent need for vector databases that manage high-dimensional vector embeddings and provide efficient vector search capabilities for AI applications. Performance, elasticity, and isolation are the key factors for vector databases to serve modern AI applications effectively. Disaggregation of storage and compute is widely recognized as the most effective approach in both academia and industry. Existing work either redesigns specialized vector databases according to the disaggregated architecture or integrates vector search into generalized databases that already use this architecture. However, challenges still remain in building elastic and efficient vector search systems within the disaggregated architecture, such as higher data fetching latency and the highly stateful nature of vector index, which hinder the system's ability to simultaneously achieve high performance, high elasticity and resource isolation. Additionally, a recent trend has emerged to integrate vector search into general-purpose databases, yet the extensibility and generality of integration methodologies have not been systematically studied. In this paper, we present BlendHouse, a cloud-native and generalized vector database system built on top of the disaggregated storage and computation architecture. BlendHouse achieves high performance, high elasticity and resource isolation simultaneously via a suite of optimizations specific to the vector search workload regarding the disaggregated architecture and the relational database. Experimental results demonstrate that BlendHouse outperforms Milvus and pgvector in terms of read and write performance. The integration methodology illustrated in this paper is extensible and general, paving the way for more powerful data management systems in the AI era.

Modern analytical query engines (AQEs) are essential for large-scale data analysis and processing. These systems usually provide numerous query-level tunable knobs that significantly affect individual query performance. While several studies have explored automatic DBMS configuration tuning, they have several limitations to handle query-level tuning. Firstly, they fail to capture how knobs influence query plans, which directly affect query performance. Secondly, they overlook query failures during the tuning processing, resulting in low tuning efficiency. Thirdly, they struggle with cold-start problems for new queries, leading to prolonged tuning time. To address these challenges, we propose AQETuner, a novel Bayesian Optimization-based system tailored for reliable query-level knob tuning in AQEs. AQETuner first applies the attention mechanisms to jointly encode the knobs and plan query, effectively identifying the impact of knobs on plan nodes. Then, AQETuner employs a dual-task Neural Process to predict both query performance and failures, leveraging their interactions to guide the tuning process. Furthermore, AQETuner utilizes Particle Swarm Optimization to efficiently generate high-quality samples in parallel during the initial tuning stage for the new queries. Experimental results show that AQETuner significantly outperforms existing methods, reducing query latency by up to 23.7% and query failures by up to 51.2%.