This sounds like a cool idea, but I don’t think it’s very usable in practice.
At least not when using SSD/HDD for storage, where — in my experience — CPU time never exceeds 50% of the time spent executing a query. This means that the maximum speedup is 2x, without even accounting for the time spent performing JIT-compilation.
So we have a situation where the maximal theoretical speedup is 2x and the maximal slowdown is unbounded.
At least not when using SSD/HDD for storage, where — in my experience — CPU time never exceeds 50% of the time spent executing a query. This means that the maximum speedup is 2x, without even accounting for the time spent performing JIT-compilation.
So we have a situation where the maximal theoretical speedup is 2x and the maximal slowdown is unbounded.