Skip to content

Latest commit

 

History

History

ydb_vector_search

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 

YDB: dealing with Big Data and moving towards AI

Speaker: Alexander Zevaykin
Slides: presentation.pdf
Event: Tech Internals Conf Cyprus 2024

YDB is a versatile open-source Distributed SQL Database that combines high availability and scalability with strong consistency and ACID transactions. It provides services for ML-products and goes ahead of vector search.

Outline

The database is used for industrial operations within Yandex. Among its clients, Yandex Market, Yandex Alice, and Yandex Taxi. They are some of the largest and heaviest AI-based applications.

The database provides true elastic scalability, capable of scaling up or down by several orders of magnitude.

At the same time, the database is fault-tolerant. It has been designed to work across three availability zones, ensuring that the database continues to function even if one of the zones becomes unavailable. The database automatically recovers after a disk failure, server failure, or datacenter failure, with minimal latency disruptions to the applications.

Currently, work is underway to implement accurate and approximate nearest neighbor search for machine learning purposes.

Takeaways

  • Architecture of a distributed fault-tolerant database.
  • Approaches to the implementation of vector search on large amounts of data.