February 2019 Summaries

In February 2019, Vespa, an open-source big data processing and serving engine primarily developed by Yahoo engineers, announced several updates aimed at enhancing its functionality and performance. The updates include the introduction of a boolean field type, which significantly reduces memory usage and increases query throughput for applications with numerous boolean fields. Additionally, Vespa now allows setting environment variables in services.xml, facilitating the use of libraries that depend on these variables, and has improved advanced search core tuning by enabling index warmup to reduce high-latency requests at startup and smoothing memory usage for growing applications. These enhancements, driven by community feedback, are part of Vespa’s ongoing development and commitment to providing robust data processing solutions.

Yahoo's implementation of an advanced comment serving system on its content sites like Yahoo Finance, News, and Sports uses reinforcement learning and the Vespa open-source platform to efficiently manage and rank large volumes of user comments. By storing comments as individual documents in Vespa, the system can handle real-time operations such as adding, finding, and displaying comments, even when facing millions of comments with varying degrees of relevance and interest. The comment ranking process involves computing a score for each comment using a ranking function configured in Vespa, which incorporates features based on user interactions and author reputation. The team improved comment selection by transitioning from a manually written ranking expression to a neural network model that leverages reinforcement learning to optimize user interest proxies like dwell time and user engagement signals. This approach has resulted in a 20% increase in average dwell time, with the neural network evaluated in parallel across content nodes to maintain low response times. Future plans include expanding this methodology to other domains such as personalized content recommendations.

Vespa.ai, an open-source big data serving solution, addresses the challenge of scaling online evaluation of machine-learned models to large datasets, a topic that will be discussed by Jon Bratseth, Distinguished Architect at Verizon Media, at the Big Data Technology Warsaw Summit on February 27th. The summit, attended by over 500 participants, focuses on big data analysis, scalability, storage, and search, featuring 27 presentations. Bratseth's talk will delve into the architecture of scalable machine-learned model serving, demonstrating how Vespa can be leveraged to efficiently serve TensorFlow and ONNX models, and will include performance benchmarks comparing Vespa to TensorFlow Serving.

Vespa 7 has been released, marking a major version update from Vespa 6, used by many high-traffic production applications on the Vespa cloud. While there are no new features in this release, as new features are introduced incrementally in minor updates, the major version change signifies the removal of deprecated legacy features and adjustments to some default settings, aligning with Vespa's practice of semantic versioning. Users planning to upgrade should review the release notes to ensure compatibility with their applications and follow the standard live upgrade procedure.