Hi Brad,
>>>>> "Brad" == Brad Dixon <[EMAIL PROTECTED]> writes:
Brad> On Thu, Mar 09, 2000 at 04:33:18AM -0500, John Michael Clemens wrote:
>> I've been put in charge of a project to use RT-Linux to do some hard
>> real-time image processing at 30-60 frames per second. What I need to do
>> is run a userspace program to analyse one image, and somehow interrupt
>> that program at every 3 ms (for 30 fps) to do some accounting and book
>> keeping, and then interrupt it every 33 ms to tell it to move onto the
>> next image. I'm at a loss for how to go about doing it.
Brad> I've not used these patches, so let me know if they work.
Brad> http://www.gardena.net/benno/linux/audio/
They work pretty well. Here we basically determined the most terrible
load we could stress our hardware with and observed the system
behavior before and after applying the patch (linux v2.2.10). It does
work pretty well, with our hardware too: Pentium 90, 48MB memory, no
sound nor video board, 2GB hard disk.
The worse load we could find consists of 200 processes in SCHED_OTHER
swapping like mad. Without the patch, our own scheduling latency test
implementation shows that any high priority thread (be it SCHED_RR or
SCHED_FIFO) doesn't get scheduled at __all__. With the patch, the
__maximum__ scheduling latency we ever observed is 35ms.
Brad> The patches were targeted for user mode audio applications
Brad> utilizing the SCHED_FIFO scheduling class. The goal was to make
Brad> sure that the audio buffer was serviced in a timely manner with
Brad> minimum latency. The page is a little scatter brained but
Brad> probably work investigating.
The patch works for SCHED_RR as well, and not only for audio type of
applications ... depending on what your requirements are, and on the
underlying hardware capabilities obviously.
We were anticipating that with our low-througput processor (P90) some
other loops in the Linux kernel would take too long and would have
required introduction of new preemption points but none showed
up. Linux with this patch applied reached the expected performance
level.
I'll gladly provide more details ... but through private Email since
this subject is a tad off topic for this list.
Gilles ([EMAIL PROTECTED])
-- [rtl] ---
To unsubscribe:
echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED]
---
For more information on Real-Time Linux see:
http://www.rtlinux.org/~rtlinux/
