FC2 devel/i386 vs VIA EPIA boards

[Lamar Owen]

On Monday 19 January 2004 03:01 pm, Leonard den Ottolander wrote:
> Ok, maybe i586 cpus work, but i386 cpus (like this Cyrix P166) do not.
> As the bug report states rpm needs non i386 instructions. I actually
> think these might be MMX instructions, so Pentium I's (before P Pro)
> might not work either.

FWIW, the PPro does not have MMX.  PII was first i686 with MMX, along with the 
larger L1 cache.

I am currently successfully running FC1 on an SMP i586 (P133, nonMMX), an SMP 
PPro 200, a UP PPro 200, a K6-2 400 (more on that one a little later), and a 
P150, nonMMX.

RPM corrupts the database on the K6-2 400 very reliably.  I found, thanks to 
the input of Barry Nathan, that the culprit was the i386 glibc's lack of NPTL 
support, and the fact that RPM's locking is less reliable when using 
linuxthreads.  Barry sent me a link to a set of glibc packages built with 
NPTL for i586, and the corruption issues went away.  I double checked it on 
the dual P133 beast, and the corruption issues I had seen there also went 
away.

In the case of the K6-2, I can cram 1.5GB of RAM on that board; the dual P133 
can work with over 1GB of RAM (it's an HP NetServer, built with a massive 
amount of RAM slots).  FC1 installed smoothly on both boxes; the P133 dually 
was previously running RH8.0.  That upgrade did not go smoothly; the box had 
to use LILO, so after the upgrade it booted into the RH8 2.4.20 kernel, which 
promptly gave thousands of scrolling error messages, but finally gave a 
login.  At that point I was glad I had a lilo binary RPM.  I then had 
duplicates of ntp andsome other packages installed, which I got rid of.   But 
the beast is running a rebuilt SMP i586 kernel now, and is pretty fast.  
Getting NPTL made a difference in the speed of many of its operations.
-- 
Lamar Owen
Director of Information Technology
Pisgah Astronomical Research Institute
1 PARI Drive
Rosman, NC  28772
(828)862-5554
www.pari.edu