October 2025 Summaries

P99 CONF 2025, hosted by Tim Koopmans, showcased a series of talks from global engineers on improving system performance, with a particular focus on reducing latency, especially P99 latency. The conference covered a wide array of topics, including modern systems programming, database and storage innovations, and the burgeoning role of artificial intelligence in vector search and large language models (LLMs). Key highlights included presentations by PayPal and TigerBeetle on managing unpredictable latency outliers, ScyllaDB's low-latency vector search engine, and discussions on performance engineering and testing. The prominence of the Rust programming language was notable, with several projects highlighted for their use of Rust to optimize performance, though Go and C++ also featured. Noteworthy presentations included Chip Huyen's keynote on optimizing LLM inference and discussions on AI-assisted analytics, underscoring the conference's focus on cutting-edge advancements in technology. The event concluded with acknowledgments to the organizers and the community, highlighting the collaborative spirit of the participants and their contributions to the field.

The blog post explores the development of a movie recommendation application that utilizes ScyllaDB's new vector search capabilities to perform semantic searches across movie plot descriptions. This innovative approach allows users to find movies based on meaning rather than keywords, using vector similarity functions such as cosine similarity, dot product, and L2 distance to compare text embeddings. The application, built with Python packages like Streamlit and Sentence Transformers, utilizes a TMDB dataset and ScyllaDB Cloud to deliver low-latency, vector-based recommendations. A detailed breakdown of the app's design includes information on the database schema, which features a vector index for efficient queries, and the process of converting user input into embeddings for similarity comparison. The post also offers guidance on setting up the app, providing source code and documentation for users interested in building with ScyllaDB Vector Search.

Amnon Heiman discusses the evolution of observability from version 1.0, focusing on metrics, logs, and traces, to Observability 2.0, which introduces context as a key component. Heiman argues for recognizing diagnostic procedures as the "fourth pillar" of observability, likening them to medical diagnostics like MRIs that are targeted, costly, and initiated based on specific hypotheses. These procedures are crucial for understanding system behavior but require explicit recognition and modeling to be effectively used by AI agents, especially in making nuanced decisions during system diagnostics. The lack of a structured approach to capturing the context, cost, and risk of these procedures could hinder the ability of AI to perform complex diagnostic tasks that humans handle intuitively. Heiman emphasizes the need for a unified description and execution framework for diagnostic procedures to enable AI to effectively match or exceed human diagnostic capabilities, thereby enhancing the overall observability framework.

ScyllaDB has introduced a low-latency Vector Search engine that is now production-ready, offering millisecond-latency vector retrieval at a massive scale, thus making it suitable for large-scale semantic search and retrieval-augmented generation workloads. The Vector Search architecture decouples vector indexing and similarity search into a dedicated Rust engine, allowing ScyllaDB nodes to pair with a local Vector Store node for optimized performance. This setup enables independent scaling of the database and Vector Store nodes, optimizing network transfer costs and allowing real-time ingestion to progress efficiently. Initial performance tests demonstrated that ScyllaDB Vector Search outperformed industry averages in both throughput and latency, sustaining high query rates with low latencies under extreme concurrency. The blog discusses various architectural design decisions, testing, and optimizations, including the use of asynchronous and synchronous threads, to enhance performance. Additionally, the blog provides insights into overcoming network-related latency challenges using techniques like disabling Nagle’s algorithm and optimizing thread layouts for better throughput and latency. The blog invites feedback from the community and encourages users to explore the Quick Start Guide for a hands-on experience.