How very thoughtful of you all to be experimenting and discussing the very
thing I am working with as well! ;-)
For reasons that I shall not elaborate on, I am very interested in
integrating a real-time flight simulation and an OpenGL accelerated visual
system (using state-of-the-art 3d acceleration hardware) in the same SMP
box. I have had a real-time flight simulation operating since last November
'99 - and have been benchmarking OpenGL apps running in X401 on top of a NMT
real-time kernel with the simulation loaded and running.
My test box is a dual 1gig PIII box with 1/2gig memory and a Cheetah hooked
to an Ultra SCSI2 controller. I have tested a variety of NVIDIA graphics
cards (Creative GeForce, GeForce II GTS, and TNT2 Ultra cards) using the
latest NVIDIA 0.9-5 GL accelerated drivers. My current benchmark involves
running Loki's UnReal Tournament GL accelerated via the NVIDIA hardware, in
one Xwindow, with the simulator running in another xwindow. With the game
running in demo mode 1024x678 res, 32 bit color, 4x4 anti-aliasing, I see
average jitter in the 10-15uec range while the game cranks out stunningly
detailed 3D world images in the 60-75 FPS range. I can move the 3D
accelerated window around and jitter remains constant.
**HOWEVER** (uh oh here comes the fine print)
I too see occasional 100-180usec jitter - it always occurs in the same place
during the demo-playback (it is a closed loop that plays on forever - or
until the box locks up - more on this in a moment). This jitter is always
preceded by an apparent momentary lock up of the user mode side (i.e. sound
stutters, screen and all windows freeze, mouse and keyboard stop responding,
etc). After what seems to be an eternity (usually 1 - 3 seconds) everything
starts back up - and my simulation debug diagnostic screen indicates that a
100-150 usec scheduling latency has occurred. The box will continue to run
and stutter at the same spot(s) although not necessarily every pass (each
pass of the demo takes about 1-2 minutes sec to play out).
If I am running a UP load, I can play the game on the internet with the
simulation running, and except for occasional momentary freezes I am able to
play normally for hours at a time. If I am running a SMP load, the whole
system will finally end up in a series of long, sustained delays, with short
momentary lapses of proper operation. My initial goal was to benchmark a box
for feasibility studies - thus my primary purpose has been achieved. But I
can't help but have a few doubts about using X4.0.1, with RTL myself.
The 150usec max jitter isn't an issue in my case since I am dealing with
periodic scheduling rates of 30hz for simulation and 60hz for visual. But
the user mode lock-ups may be reason for genuine concern. My assumption up
to now was that there was non RT/SMP code in the UnReal game itself causing
these problems (probably related to disk access involved in reading the 3D
environment off the disk during display). Since I have access to some Linux
based GL accelerated flightsim game source, and reasonable benchmarks, I
decided to trudge on to the next level of my tests (which involve extracting
the visual portion of the flightsim source and integrating it with my
real-time code).
I have also been conducting tests under X4.0.1 and my real-time simulation
using NON GL accelerated tasks. Including moving a normal window about,
running Netscape, network access, etc. Operation is SOLID and reliable under
these conditions with occasional (once every 10 minutes or so) jitter in the
80usec range (which appears to be more network related). As a matter of fact
I monitored election returns (a totally futile undertaking) real-time on the
Net while my sim ran. During this time I had SEVERAL instances of Netscape
open (one to each major network). Everything worked perfectly - and left me
with a nice "warm fuzzy feeling" about Linux, X, and NMT (more than I can
say about the election!). I have also done a lot of testing/observation on
boxes at work (these other machines are personal home boxes). All different
configurations, and generations of processors and video cards. The 80usec
rare occasional jitter is a very common phenomena on all boxes I have tested
to date.
Anyway you expressed interest in any/all such projects - for now this is my
2 cents worth...
Best Regards,
Todd Gearheart
-- [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/
