Re: [2.4.17/18pre] VM and swap - it's really unusable

yodaiken@fsmlabs.com
Mon, 21 Jan 2002 14:54:38 -0700


On Mon, Jan 21, 2002 at 04:22:58PM -0500, Robert Love wrote:
> On Mon, 2002-01-21 at 11:50, yodaiken@fsmlabs.com wrote:
> > On Mon, Jan 21, 2002 at 05:48:30PM +0100, Daniel Phillips wrote:
>
> > > Consider a thread reading from disk in such a way that readahead is no help,
> > > i.e., perhaps the disk is fragmented. At each step the IO thread schedules a
> > > read and sleeps until the read completes, then schedules the next one. At
> > > the same time there is a hog in the kernel, or perhaps there is
> > > competition from other tasks using the kernel. In any event, it will
> > > frequently transpire that at the time the disk IO completes there is somebody
> > > in the kernel. Without preemption the IO thread has to wait until the kernel
> > > hog blocks, hits a scheduling point or exits the kernel.
> >
> >
> > So your claim is that:
> > Preemption improves latency when there are both kernel cpu bound
> > tasks and tasks that are I/O bound with very low cache hit
> > rates?
> >
> > Is that it?
> >
> > Can you give me an example of a CPU bound task that runs
> > mostly in kernel? Doesn't that seem like a kernel bug?
>
> It doesn't have to run mostly in the kernel. It just has to be in the
> kernel when the I/O-bound tasks awakes. Further, there are plenty of

How does that work? Won't the switch happen on exit from the kernel?

> what we consider CPU-bound tasks that are interactive and/or
> graphics-oriented and this adds much to their time in the kernel.

I'm not sure what an "interactive and/or graphics-oriented" CPU bound
task might be. Is there a definition?

> In a given period of time, a CPU bound task can run at any allotment
> within it is given. On the other hand, an I/O-bound task spends much
> time blocked and thus can only run when I/O is available and it is
> awake. It is thus advantageous to schedule it within the bounds of the
> I/O being available, and as tightly in those bounds as possible. This
> more fairly distributes scheduling to all tasks. Same goes for RT
> tasks, interactive tasks, etc.

So you think of an "I/O bound task" as "an I/O bound task that spends
most of its timeblocked". Won't the latencies of such tasks already be
pretty high? I'd think that better caching and read-ahead is the correct
fix.

> The result is faster wake-up-to-run and thus higher throughput. I just
> sent some dbench scores to correlate this.
>
> > I still keep missing these reports. Can you help me here?
> > (Obviously "my laptop seems more effervescent" is not what I'm looking
> > for.)
>
> While we certainly need tangible empirical benefits, users finding their
> desktop experience smoother and thus more enjoyable is just about the
> best thing we can ask for.

It depends on what you want.

-- 
---------------------------------------------------------
Victor Yodaiken 
Finite State Machine Labs: The RTLinux Company.
 www.fsmlabs.com  www.rtlinux.com

- 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/