[QUESTION]: Why does ORM-style API seem to outperform MilvusClient in database search operations?

[cococo2000]

Is there an existing issue for this?

• I have searched the existing issues

What is your question?

I installed milvus (v2.4.1) standalone with docker compose following https://milvus.io/docs/install_standalone-docker-compose.md. and I tried to test milvus performance with mnist dataset , HNSW index and L2 distance.

However, I found that the ORM-style API seems to outperform MilvusClient in database search operations.

First, I followed the Hello-Milvus tutorial and hello_milvus.py in milvus-docs using MilvusClient.

from pymilvus import MilvusClient, DataType

client = MilvusClient(
    uri="http://localhost:19530"
)

...

But, I found that the QPS was only around 4. Then, I tried to use ORM-style API following hello_milvus.py in pymilvus and the QPS was over 150 which is much higher than MilvusClient.

from pymilvus import (
    connections,
    utility,
    FieldSchema, CollectionSchema, DataType,
    Collection,
)

connections.connect("default", host="localhost", port="19530")

collection_name = "test"
if utility.has_collection(collection_name):
    utility.drop_collection(collection_name)
...

I'm not sure why ORM-style API runs much faster than MilvusClient. Could you provide some insights on this?

Anything else?

No response