October 2023 Summaries

The October 2023 Vespa Newsletter highlights significant updates and improvements in Vespa.ai, including the launch of an enhanced search UI with AI-generated suggestions and paragraph indexing, as well as the integration of Lucene Linguistics, which supports 40 languages. Vespa.ai has become an independent company, aiming to deliver features more rapidly and provide cost-effective, high-quality solutions on Vespa Cloud, which now offers Vespa Cloud Enclave for enhanced data control. Notable performance advancements include optimizations in fuzzy matching using Deterministic Finite Automata, resulting in substantial speed improvements, and the introduction of automated reindexing for BM25 ranking. Other updates include cluster-specific model-serving settings, enhanced deployment status tracking in the Vespa CLI, and performance improvements in handling small allocations for paged attributes. Additionally, Vespa's use of ONNXRuntime for model inference now supports bfloat16 and float16 datatypes. The newsletter invites users to try Vespa's capabilities by deploying applications for free on Vespa Cloud.

The blog post introduces the new Lucene-based Vespa Linguistics component, which aims to facilitate the migration of search applications from other Lucene-based search engines like Elasticsearch to Vespa, while also improving existing Vespa applications. The component, developed after insights gained at the Berlin Buzzwords 2023 conference, leverages Apache Lucene's capabilities to support 40 languages and offers a customizable approach to text analysis. It simplifies the migration process by allowing for the conversion of Elasticsearch analyzers to Vespa configurations, minimizing the need for extensive overhauls. The post highlights the flexibility Vespa offers over Elasticsearch, particularly in deploying and maintaining custom text analysis components without the need for plugins. This flexibility not only reduces the complexity associated with version management and deployment but also ensures that those familiar with Lucene-based technologies can quickly adapt to using Vespa.

On October 10, 2023, a significant vulnerability known as HTTP/2 Rapid Reset (CVE-2023-44487) was disclosed, affecting numerous HTTP/2 servers, including those using Vespa's technology through Jetty. To address this issue, Jetty released version 11.0.17, and Vespa promptly built and deployed version 8.240.5 to Vespa Cloud, ensuring that users of Vespa Cloud were immediately protected without needing to take further action. However, self-hosted users are urged to upgrade to this latest version to safeguard their systems. For further assistance, Vespa provides support through their Slack channel, and additional reading is available on Google's security-focused blog posts, which discuss the mitigation of large-scale DDoS attacks, including the HTTP/2 Rapid Reset attack.

Vespa.ai is transitioning from being a project within Yahoo to becoming an independent company, focusing on leveraging AI for big data applications at scale. Originally developed to address Yahoo's needs in search, recommendation, and ad serving, Vespa has grown since its open-sourcing in 2017 to become a preferred platform for enterprises dealing with large language models and vector databases. To enhance scalability and efficiency, a centralized cloud service was developed, significantly reducing resource usage and freeing employee time while maintaining high-quality operations. Despite its independence, Vespa will maintain a relationship with Yahoo, which will remain a major customer and stakeholder. The new company aims to extend its cloud service efficiencies and advanced AI capabilities globally, facilitating the development of innovative online AI applications, while continuing to provide an open-source option for users.