Re: [PATCH 0 of 4] Generic AIO by scheduling stacks

[Linus Torvalds <torvalds@linux-foundation.org>]

On Sat, 10 Feb 2007, Linus Torvalds wrote:
> 
> But that makes it impossible to do things synchronously, which I think is 
> a *major* mistake.
> 
> The whole (and really _only_) point of my patch was really the whole 
> "synchronous call" part. I'm personally of the opinion that if you cannot 
> handle the cached case as fast as just doing the system call directly, 
> then the whole thing is almost pointless.

Side note: one of the nice things with "do it synchronously if you can" is 
that it also likely would allow us to do a reasonable job at "self-tuning" 
things in the kernel. With my async approach, we get notified only when we 
block, so it'seasy (for example) to have a simple counter that 
automatically adapts to the number of outstanding IO's, in a way that it's 
_not_ if we do things at submit time when we won't even know whether it 
will block or not.

As a trivial example: we actually see what *kind* of blocking it is. Is it 
blocking interruptibly ("long wait") or uninterruptibly ("disk wait")? So 
by the time schedule_async() is called, we actually have some more 
information about the situation, and we can even do different things 
(possibly based on just hints that the user and/or system maintainer gives 
us; ie you can tune the behaviour from _outside_ by setting different 
limits, for example).

			Linus