[svlug] upgrading to a real O/S

[Bryan-TheBS-Smith]

Ian MacLure wrote:
> In my current work env I ( and others ) are lumbered with
> that alternative O/S (the 2K version).  The app we are
> sweating over is aerospace related ( multiple high fidelity
> flight sims networked with some high end automated traffic
> avoidance type s/w and target generation apps ).  We are
> currently running all this on a bunch of dual 1G/512M intel
> PIII ( soon to be 1.7G PIV ) boxes

Er, for heavy FPU work and a lot of I/O, I'd seriously consider
dual-Athlon MP.  The Athlon core excells at FPU compared to a P4
(heck, a P3 running at half the speed is faster than a P4 at
double-precision floating point!) and the cross-bar interconnect of
the Alpha EV7 bus used by the Athlon allows more than two nodes to
transfer data simultaneously (unlike the Intel SMP approach).

Look at these benchmarks to see what I mean (note those P4's are
costly _Xeons_!):
http://www.anandtech.com/chipsets/showdoc.html?i=1483&p=14
http://www.anandtech.com/chipsets/showdoc.html?i=1483&p=15
http://www.anandtech.com/chipsets/showdoc.html?i=1483&p=16
http://www.anandtech.com/chipsets/showdoc.html?i=1483&p=17

> with the graphics heavy lifting done by 3D Lab Wildcat II
> 5110 boards driving dual monitors (at least for the flight
> sim part). We also have flat screen repeaters coming off
> the same boxes.

I don't know of any 3D Labs drivers for Linux, but you may want to
check out nVidia's Quanta2 chip-based solutions for high-end,
dual-headed OpenGL on Linux.  SGI might be the first vendor to
consider.  On the cheap, you can go with nVidia's GeForce2 MX which
can also do dual-head, but I don't think it'll give you the
performance you desire.  The Quanta3, which will be based on the
GeForce3, should be out shortly.

> What I'd like to do is implement this on a real O/S (nudge-
> nudge ) and was wondering if there is anybody out there with
> experience of running the 3D boards with Linux ( of whatever
> flavo(u)r ). This would be the Achilles heal of the whole
> effort and a show stopper if board perf under Linux was less
> than on that other (P)OS.  It would help if graphics performance
> were better on Linux actually but I'm not holding my breath.
> I'll settle for no worse.

Only nVidia offers compelling Linux support.  Their unified driver
model uses the same codebase for Linux and Windows.  As such,
performance is just about on-par between the two.  It is the only
solution I would consider.  Their newer drivers can drive OpenGL
across two monitors as one frame buffer.

> Code is C++ ( and FORTRAN [ but not where the graphics matter ]
> ) with OpenGL doing the graphics stuff. We are sort of stuck
> with OpenGL ( which isn't so bad considering the alternative )
> for now.  Any advice/experience you can share?

Yes.  At a previous employer, also in aerospace, I built a Linux
cluster that did all the computations on the backend, and then sent
GLX (basically OpenGL over X-Windows) 3-D to the front-end. 
Basically, no Windows solution can match our rendering throughput in
real-time, and we also used the cluster for non-real-time
computational fluid dynamics (CFD) simulations on off-hours.

Intel has also released its optimizing C/C++/Fortran commercial
compilers for Linux.  And the standard GNU Toolchain offers
C/C++/Fortran77 compilers as well.

> Thanx.

BTW, we have an engineer with a firm here in Orlando who uses the
nVidia-Linux combination for low-cost (i.e. dual-processor PC-based,
just like yourself) flight simulation systems.  His name is Darren
Humphreys of DISTI.  I'll give you his E-mail address when I get
home (I don't have it here).

-- TheBS

-- 
Bryan "TheBS" Smith   mailto:b.j.smith at ieee.org    chat:thebs413
Engineer  AbsoluteValue Systems, Inc.  http://www.linux-wlan.org
President    SmithConcepts, Inc.    http://www.SmithConcepts.com