Linus Torvalds wrote:
>
> ...
> A fast disk driver will _never ever_ do a partial request completion. A
> high-performance subsystem will put in the scatter-gather list and say
> "go" to the controller, and the controller will send exactly one interrupt
> back when it is all done.
OK. But still, I don't see why we need partial BIO completions. If
we say that the basic unit of completion is a whole BIO, then readahead
can then manage latency via the outgoing bio size.
> So for such a system, you'd never see partial completions anyway.
>
> Partial completions are a feature of slow hardware. And slow hardware is
> exactly when we want to know about it.
Well I'd be interested in knowing specifically what is wrong with the
behaviour of 2.5.33 against a floppy disk.
In the testing I did a few weeks back, everything checked out. An
application which was reading the raw device at 95% of media bandwidth
never blocked. An application which was capable of processing data at
120% of media bandwidth achieved 100%.
It could be that the initial 64k read at the start-of-file is
too big, and the many-small-file behaviour is poor?
A specific "this sucks" testcase would be helpful...
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Sep 07 2002 - 22:00:26 EST