Re: stochastic fair queueing in the elevator [Re: [BENCHMARK] 2.4.20-ck3 / aa / rmap with contest]

[Andrea Arcangeli]

On Mon, Feb 10, 2003 at 11:36:50PM +1100, Nick Piggin wrote:
> in corner cases. Luckily it seems you don't need such
> infrastructure to demonstrate most anticipatory scheduler gains.

sure. I guess it would be most important to be allowed to leave
anticipatory scheduling enabled *always* even with very async-io
intensive workloads, so when you read a file with normal fileutiles you
still can take advantage of it for the most important places. Ideally in
this case where w/o the hint we might have to disable anticipatory
scheduling completely to get peak async-io performance, the non-hint
timeout could be zero (i.e. no stall at all) and the kernel would
understand and do the right thing

Andrea