The Henry Spencer Effect

On Sep 18, 1:17 pm, kT wrote:
Hi Everybody!

I was really surprised how quickly I was able to identify and reproduce
the Henry Spencer Effect, after just a couple of weeks into the project.

For those of you unfamiliar with this amazing effect, it's the breakdown
of obsolete hardware and software systems under the burden of operating
at modern standards. Generally, it is manifested in the overloading of
the DMA channel, resulting in the corruption of files during large file
transfer, usually when multitasking, with the inevitable tragic results.

The K6 Socket 7 motherboards of the 66 MHz FSB variety with the highly
integrated chipsets are the most problematic, especially when the bus
chips also integrate video, but the Super Socket 7 100 Mhz FSB systems
are much more immune to this. For instance, this system is a very old
K6-2 3DNow system, and the only problem I'm having is very slow keyboard
interrupt response when running continuous cable radio Jazz streaming.
Most likely the keyboard problem will disappear if I upgrade to a modern
audio card and drivers, and replace the SCSI bus controller and SCSI
drive with the IDE drive for which this system was originally designed.

Yesterday, I actually got Orbiter to run on a 500 MHz K6-2 3DNow box. It
was a real mess to find an nVidia driver capable of accomplishing that,
and I still have to tweak the driver version to get it to be stable, but
that is a major accomplishment in my book of old dinosaur experiments.

Thanks Henry! You're a genius.

Turn it into a Linux box (but an older version like RedHat 7.2) and be
done with it.

Eric

Hi Everybody!

I was really surprised how quickly I was able to identify and reproduce
the Henry Spencer Effect, after just a couple of weeks into the project.

For those of you unfamiliar with this amazing effect, it's the breakdown
of obsolete hardware and software systems under the burden of operating
at modern standards. Generally, it is manifested in the overloading of
the DMA channel, resulting in the corruption of files during large file
transfer, usually when multitasking, with the inevitable tragic results.

The K6 Socket 7 motherboards of the 66 MHz FSB variety with the highly
integrated chipsets are the most problematic, especially when the bus
chips also integrate video, but the Super Socket 7 100 Mhz FSB systems
are much more immune to this. For instance, this system is a very old
K6-2 3DNow system, and the only problem I'm having is very slow keyboard
interrupt response when running continuous cable radio Jazz streaming.
Most likely the keyboard problem will disappear if I upgrade to a modern
audio card and drivers, and replace the SCSI bus controller and SCSI
drive with the IDE drive for which this system was originally designed.

Yesterday, I actually got Orbiter to run on a 500 MHz K6-2 3DNow box. It
was a real mess to find an nVidia driver capable of accomplishing that,
and I still have to tweak the driver version to get it to be stable, but
that is a major accomplishment in my book of old dinosaur experiments.

Thanks Henry! You're a genius.

Eric Chomko wrote:
On Sep 18, 1:17 pm, kT wrote:
Hi Everybody!

I was really surprised how quickly I was able to identify and reproduce
the Henry Spencer Effect, after just a couple of weeks into the project.

For those of you unfamiliar with this amazing effect, it's the breakdown
of obsolete hardware and software systems under the burden of operating
at modern standards. Generally, it is manifested in the overloading of
the DMA channel, resulting in the corruption of files during large file
transfer, usually when multitasking, with the inevitable tragic results.

The K6 Socket 7 motherboards of the 66 MHz FSB variety with the highly
integrated chipsets are the most problematic, especially when the bus
chips also integrate video, but the Super Socket 7 100 Mhz FSB systems
are much more immune to this. For instance, this system is a very old
K6-2 3DNow system, and the only problem I'm having is very slow keyboard
interrupt response when running continuous cable radio Jazz streaming.
Most likely the keyboard problem will disappear if I upgrade to a modern
audio card and drivers, and replace the SCSI bus controller and SCSI
drive with the IDE drive for which this system was originally designed.

Yesterday, I actually got Orbiter to run on a 500 MHz K6-2 3DNow box. It
was a real mess to find an nVidia driver capable of accomplishing that,
and I still have to tweak the driver version to get it to be stable, but
that is a major accomplishment in my book of old dinosaur experiments.

Thanks Henry! You're a genius.

Turn it into a Linux box (but an older version like RedHat 7.2) and be
done with it.

I couldn't even get Debian to run on it. And I went through multiple
Socket 7 boards to confirm what I was actually seeing. I couldn't get a
single one of them to work properly on anything. I am done with it.

I don't need a Linux firewall that bad. On the other hand, I have a
couple of these super socket 7 systems (PC-100) that are upgradeable to
K6-2+ and K6-III, and there is still a large community of these users
out there, so getting Orbiter to run on such boxes would be a major
improvement. I only got half way through the process yesterday (it was
the 150th anniversary of the birth of Konstantin Tsiokovsky, so we were
busy with other things) but we got almost all the way to orbit at 10 FPS
with a system with no internal second level cache, before it crashed.
That could have been simple overheating of the CPU, who knows. It was a
Windows 2000 system with DirectX 9.0c, an MX-4000 in a PCI slot, and I
had to revert back to nVidia driver 5x.xx to make it work at all :

http://www.nvidia.com/object/winxp-2k_archive.html

That's a very long list of drivers to sort through.

kT wrote:

Eric Chomko wrote:
On Sep 18, 1:17 pm, kT wrote:
Hi Everybody!

I was really surprised how quickly I was able to identify and reproduce
the Henry Spencer Effect, after just a couple of weeks into the project.

For those of you unfamiliar with this amazing effect, it's the breakdown
of obsolete hardware and software systems under the burden of operating
at modern standards. Generally, it is manifested in the overloading of
the DMA channel, resulting in the corruption of files during large file
transfer, usually when multitasking, with the inevitable tragic results.

The K6 Socket 7 motherboards of the 66 MHz FSB variety with the highly
integrated chipsets are the most problematic, especially when the bus
chips also integrate video, but the Super Socket 7 100 Mhz FSB systems
are much more immune to this. For instance, this system is a very old
K6-2 3DNow system, and the only problem I'm having is very slow keyboard
interrupt response when running continuous cable radio Jazz streaming.
Most likely the keyboard problem will disappear if I upgrade to a modern
audio card and drivers, and replace the SCSI bus controller and SCSI
drive with the IDE drive for which this system was originally designed.

Yesterday, I actually got Orbiter to run on a 500 MHz K6-2 3DNow box. It
was a real mess to find an nVidia driver capable of accomplishing that,
and I still have to tweak the driver version to get it to be stable, but
that is a major accomplishment in my book of old dinosaur experiments.

Thanks Henry! You're a genius.

Turn it into a Linux box (but an older version like RedHat 7.2) and be
done with it.

I couldn't even get Debian to run on it. And I went through multiple
Socket 7 boards to confirm what I was actually seeing. I couldn't get a
single one of them to work properly on anything. I am done with it.

Try a different distribution

http://distrowatch.com/

DamnSmallLinux comes to mind, but you've already chosen the best option.


I don't need a Linux firewall that bad. On the other hand, I have a
couple of these super socket 7 systems (PC-100) that are upgradeable to
K6-2+ and K6-III, and there is still a large community of these users
out there, so getting Orbiter to run on such boxes would be a major
improvement. I only got half way through the process yesterday (it was
the 150th anniversary of the birth of Konstantin Tsiokovsky, so we were
busy with other things) but we got almost all the way to orbit at 10 FPS
with a system with no internal second level cache, before it crashed.
That could have been simple overheating of the CPU, who knows. It was a
Windows 2000 system with DirectX 9.0c, an MX-4000 in a PCI slot, and I
had to revert back to nVidia driver 5x.xx to make it work at all :

http://www.nvidia.com/object/winxp-2k_archive.html

That's a very long list of drivers to sort through.

Yeah, old hardware, old concepts.

Speaking of Nvidia and video cards, here is something new (I've been waiting
for ten years for one of these, didn't notice their arrival last year. But
here they are :-)

How does your Orbiter run with one of theses?
http://www.nvidia.com/page/geforce_8800.html

Want to use it for more than just playing games? Well, then add this.
http://developer.nvidia.com/object/cuda.html

And you get this,
http://www.youtube.com/watch?v=WCU7wlusSDI
Great video ;-)

The N-Body problem with N = 131072 at 346 Gigaflops.

--
Craig Fink
Courtesy E-Mail Welcome @