if you want high throughput want to optimize the every component in the pipeline i try the dragonflydb pretty good comparing other database also if you add reranking like methods the ms gonna high