Re: [PATCH] scheduler patch, faster still

Jamie Lokier (lkd@tantalophile.demon.co.uk)
Wed, 30 Sep 1998 09:48:21 +0100


On Tue, Sep 29, 1998 at 04:11:56PM -0600, yodaiken@chelm.cs.nmt.edu wrote:
> One easy project is to support memory managed RTLinux tasks. This
> was the original design and we dropped it for performance reasons and
> when modules started working well. But it should be relatively easy to
> take a 4Meg chunk of memory, reserve it for RTLinux modules, make
> a map for it, and then fix the rtscheduler to really switch.

Though for the kinds of things I'm doing, we use threads a lot in a
single main process anyway. There are context switches (though we're
investigating keeping moving that mostly to the user-space thread
library).

The kernel context switching issue is still relevant because (a) other
control processes, general handy daemons and the like and the odd shell
are still running occasionally; (b) I'm personally not convinced that
polling DMA'd system memory throughout the user-space code (U-Net style)
is much better than using hardware interrupts, and it certainly
complicates the application, so I'd like to see the kernel do a good job
itself.

-- Jamie

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/