When running Milvus 2.0 ga standalone using docker compose, what is the level of performance that can be expected for a collection of 100M 256-dimensional vectors (with an IVF_SQ8 index, nlist = 65536), in terms of # queries/second? The current performance I observe seems very slow (cannot surpass a peak of 1 req/second in the very best case, on a very large EC2 server with abundant cpu and memory). Is a max of 1 request/second the expected performance for a Milvus 2.0 standalone instance with a collection like mine, or should it be much faster?
Asked
Active
Viewed 283 times
1 Answers
0
The official community has released the benchmark
https://milvus.io/docs/v2.0.x/benchmark.md#Milvus-20-Benchmark-Test-Report
coolflower
- 69
- 8