What's new in Milvus 2.1 - Towards simplicity and speed

What this post added

Milvus 2.1 introduces significant performance enhancements including a new routing protocol reducing retrieval latency to 5ms, a new cost evaluation model and concurrency scheduler for a 3.2x performance boost with small-nq queries, and in-memory replicas for improved scalability and availability. Data loading is accelerated by Zstandard compression and goroutine pools. String data types (VARCHAR) are now supported as scalar data, usable as primary keys and for attribute filtering, with MARISA-Tries used for scalar inverted indexes to improve filtering speed. Kafka is added as an alternative message queue to Pulsar. A production-ready Java SDK is released, and observability is improved with new monitoring metrics and optimized log keeping. Embedded Milvus is introduced for simplified local development.