2.1.120-pre3 on IBM netfinity 7000 - impressive

Hubert Mantel (mantel@suse.de)
Fri, 4 Sep 1998 18:13:10 +0200


Hi,

I have a nice toy here and would like to report some experiences (and one
problem):

The machine is an IBM netfinity 7000 server. 4 PPro200 processors, 2 GB
RAM, two aic7880 controllers, EEpro100 network card.

First of all: Up to now the system is running rock solid under
2.1.120-pre3. We're just starting stress testing.

Here the problems and the workarounds:

Ethernet adapter always got IRQ 9. Moving to another slot gave it IRQ 10.
Standard driver in 2.1.120 is 0.36 which could not handle the card. Update
to 1.03 fixed the problem (btw 1.03 cannot be compiled into the kernel;
it's only compilable as module).

The second aic7880 (currently completely unused) got IRQ 9, too. The
aic7xxx driver (tested 5.0.20 and 5.1-pre8) ran into a SCSI timeout so the
machine could not boot. I modified the driver to ignore the second adapter
and there were no problems afterwards (currently running 5.1-pre8 with a 4
Gig system disk and two 4 Gig disks as one raid0 test device).

After increasing NR_TASKS to 2000 and /proc/sys/fs/file-max to 16384
(/proc/sys/fs/inode-max to 49152) I was able to compile 2.1.120-pre3 using
"make -j". Load went up to 170 but the compile completed successfully.
You need about 800 MB RAM to compile a kernel with "make -j".

No problems with X so far.

One problem left: Currently the kernel only uses about 980 MB of memory. I
would like to use the second Gig, too. I modified include/asm/page.h and
arch/i386/vmlinux.lds according to the comments in these files. But this
kernel is unable to boot. The last message is

Sep 4 14:53:53 Test100-DHCP9 kernel: Freeing unused kernel memory: 44k freed

After that, the system freezes. Is there anything more to be done to use 2
Gig of RAM? Should this work with the current kernel or am I doing
something wrong? Any pointers to additional docu?

The kernel looks really quite stable right now. We just completed a big
database test with about 350 MB of shared memory. Now putting additional
load to the machine and see how it goes...
-o)
Hubert Mantel /\\
_\_v

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/faq.html