Description
I ran Weaviate locally using its binary distribution and employed the Vector Database Benchmark Tool (VectorDBBench) to evaluate performance. After loading data with the command:
vectordbbench weaviate --url http://xx.xx.xx.xx:8080 --case-type Performance768D10M --db-label weaviate_test --no-auth --ef 512 --num-concurrency 96
I executed multiple test iterations with the --skip-load
parameter to bypass data reloading. Despite identical configurations, the observed Queries Per Second (QPS) exhibited significant volatility across runs (e.g., 1900, 1700, 2000, 1600). What factors could contribute to this performance inconsistency?
Server Setup Information
- Weaviate Server Version: 1.31.1
- Deployment Method:
- Multi Node? Number of Running Nodes:
- Client Language and Version:
- Multitenancy?: