[lug] Asynchronous I/O for LInux kernal

[Jonathan Corbet]

> > I just read from newsforge that the Linux kernal is switching to
> > asynchronous I/O.
> > This is probably a good thing, but not sure what it all means.
> 
> It's more efficient, perhaps more responsive/lower latency. Programs
> would have to be written to take care of it, though the program might be
> the device driver (in which case everything using the device driver
> would be taking advantage).

With all humility, LWN has more and better information on this decision
than NewsForge does...

The asynchronous switch will be internal to the kernel.  No applications
will notice the difference - unless, of course, they use the new asynch I/O
API.  Drivers will have to change, of course, though even that is somewhat
optional for noncritical drivers.  Linus's position was that it's OK for
even an explicitly asynchronous operation to block on occasion - for
example, when it's going to a device whose driver does not support
asynchronous operations.

The cases that really matter, most of the time (block devices, networking)
are already done asynchronously within the kernel now.

It's a good thing, but most Linux users will never really know the
difference.

Greetings from Ottawa,

jon

Jonathan Corbet
Executive editor, LWN.net
corbet at lwn.net