From: Angus Dorbie (dorbie++at++sgi.com)
Date: 01/24/2000 14:23:06
KOHARCHIK++at++mcopn1.rsc.raytheon.com wrote:
>
> Hi folks,
>
> We're looking for a way to draw a panoramic (120 degrees FOV) on
> Performer using Linux. Our current system uses three monitors
> positioned side by side in an arc, each displaying 40 degrees. If we
> could drive three (or four) monitors and keep things reasonably well
> synchronized, I think that would do the job. (Although I'm open to
> suggestions about other display hardware too.)
>
> I have heard some discussion about synchronizing multiple hosts
> over UDP to say what to draw. I gather the consensus is that this
> approach introduces too many artifacts to be satisfactory?
>
> I've also seen a post about tying pins from different vga
> connectors on differnt hosts together to coordinate vertical redraw.
> Did I understand that right? Has anyone done it? I can see how that
> would tell you when to draw, but how does that help you with what to
> draw ? (Eg something on your scene may have changed and been picked up
> by one host and not the other ones it's tied to by VGA sync so you end
> up drawing different stuff, but at the same time.
>
> One thing I haven't heard yet is talking about multiple graphics
> cards per host. Can Performer or the Xserver split the scene over
> several cards? Are there (I guess they'd have to be PCI) graphics
> cards that have hardware acceleration for OpenGL?
Several problems, the first is multiple head with multiple gfx card
support in Linux which isn't quite there yet.
The second is raw performance, a multihead system is going to be slower
than single head, not least of the reasons being you only have one AGP
slot, but even if you found PCI bus sufficient you'd be contending for
bandwidth on the bus unless you had multiple PCI subsystems.
The third problem is Performer's current lack of SMP (symmetric
multiprocessing) support on Linux. Performer forks draw processes to
efficiently support multipls graphics subsystems. Currently this is not
possible and won't be until SGI releases the SMP version of Performer.
>
> One approach I've heard about is a program called MetaVR. I don't
> know any details but my understanding is that they chain their systems
> together through the parallel port to stay synched. Does anyone know
> how that works?
Think about it, the vertical retrace needs video hardware, using the
parallel port isn't going to change that. It IS possible, but the real
solution is to look for a card with frame lock or get your soldering
iron out. Once you find that then you can implement whatever software
locks you like over the connection between the machines.
Everyone is in the same boat when it comes to the synching issues, IHV's
have control of that not ISV's, so the problems already discussed are
still applicable irrespective of the communication mechanism between
machines.
Cheers,ANgus.
--
"It takes a vision to beat a vision."
-Thomas Sowell
This archive was generated by hypermail 2b29 : Mon Jan 24 2000 - 14:23:20 PST