> Read with O_DIRECT apparently the challenge explicitly allow for warming up the os cache on a prior run, so O_DIRECT would be disadvantaged here as the worki

> Read with O_DIRECT apparently the challenge explicitly allow for warming up the os cache on a prior run, so O_DIRECT would be disadvantaged here as the working set fits in memory. edit: also in my very simple tests a good hashmap can beat linear search already at <10 items. It depends a lot on the hash function cost.