August 2022 Summaries

Pinecone, a company specializing in vector databases, has undergone significant developments and transformations, driven by both technical challenges and the need for innovation. Initially relying on graph-based techniques for vector search, Pinecone faced the complexities of integrating these methods into a production-grade database, requiring efficient metadata filtering and optimized updates. The limitations of using RocksDB for storage led to the development of a new vector store, "memkey," which enhanced performance and reduced operational costs. Simultaneously, the company transitioned from a C/C++ and Python codebase to Rust, despite the risks associated with rewrites and the team's unfamiliarity with the language, ultimately achieving improved performance, development velocity, and operational stability. Pinecone continues to explore unanswered questions about vector search and remains committed to advancing its technology while inviting collaboration and feedback from its user community.

Pinecone has announced significant updates to its vector database platform, enhancing its capabilities for developers to implement vector search in their cloud applications more efficiently and cost-effectively. The new features include vertical scaling, which allows for index capacity expansion without downtime, and collections that provide centralized storage and experimentation for vector embeddings and metadata. The introduction of p2 pods offers improved performance, achieving up to 10x faster search speeds for high-throughput applications, while also supporting filtering and live index updates. Additionally, performance improvements for s1 and p1 pods result in 50% lower latency and higher throughput, with the new pricing structure going into effect for new users starting September 1, 2022. Existing users will retain their current rates, and the Starter plan now includes 5x more storage capacity with s1 pods, enabling users to explore the benefits of vector search more thoroughly.