Page History

Image Added

Stability improvement

Stability is the first priority of milvus for now.

        Implement Mini Milvus Cluster and add more e2e test in the repo.

        Clean all the chaos test issue and ensure all failure can be recovered in less than 3 Minutes.

        Improve the availability with multiple replicasThe community decide to improve stability to a new step in the next major release, including:

Performance

       2C8G 1M 128 dim Vector, serial search latency < 5ms

       2C8G 1M 128 dim Vector, concurrent search latency < 50ms

       Support large cluster with tens of billion vectors and scale linearly

Major functionalities 

        Partition refine

              1. Support dynamically load/release one partition

              2. Manage the memory when many partitions are created.

              3. Improve partition search performance

        Data backup

        Resource Group

              1. physical isolation between resource groups

              2. resource group can be also combined with a namespace.

       Range Search 

       Rolling Upgrade/Graceful shutdown

       Change Data capture support

SDK

          Go SDK fully tested

...

1. Refactoring Data storage format to support new scalar execution engine and better compression
2. Refactoring data coordinator and index coordinator, combine to single process and make it easy to understand and deploy
3. Refine QueryNode V2 to simplify search path and improve search performance 

Ecosystem

1. Support Huggingface , PaddlePaddle and OpenAI as the inference engine 
2. Support Data Migration from Milvus 1.0, Faiss, Hnswlib to Milvus 2.0
3. Support to deploy Milvus on GCP easily
4. Feder support - Vector data visualization
5. More trouble solving tools such as tracing, birdwatcher and dynamic config support

Documents