[the8472]

Those are rarely the slow ones though. Lots of software simply has not been written to keep IO queues full. They read a bit of data, process it, read the next bit and so on. On a single thread. This makes all kinds of IO (including network) way slower than it has to be.

For example a tree-index can be parallelized by walking down different branches. On top of that one can issue a prefetch for the next node (on each branch) while processing the current ones.

[ilyt]

Yup, a lot of software is (was?) written with assumptions that mattered with spinning rust. And even if author didn't intend to, serial code generates serial, dependent IO.