On Mon, 2 Nov 1998 [EMAIL PROTECTED] wrote:
> I am trying to check out the SMP stability of the development and stable
> kernels. I would like to answer the question: "Can I put an enterprise database
> on a SMP Linux machine?" I try to get an answer simulating a reasonable
> CPU+DISK+MEMORY load by running kernel compile (minimal configuration,
> building zImage only) 100 times in a row. No crazy `make -j', nor even a
> moderate `make -j 4' (yet!), I am doing a simple make zImage:
make -j will now "blow up". IE it used to just do makes on each
directory, firing off one process per compile. No longer. When I
try this now I get "no more process" errors as it will quickly load up
500 compiles at once... :)
> -----------------------------------
> #!/bin/sh
> for i in 0 1 2 3 4 5 6 7 8 9
> do
> for j in 0 1 2 3 4 5 6 7 8 9
> do
> make dep;make clean
> make zImage >log.$i.$j 2>&1
> done
> done
> ------------------------------------
> Mind it, the load never goes beyond 2 and it is below 1.5 most of the time.
>
try -j2, then -j3, then -j4 and so forth. You can ramp the load and I/O
up as far as you want by adding that...
>
> Once this test succeeds (machine does not lock up) I would like to run
> parallel `make -j 4' 100 times in a row. The probelem is that it does not
> succed now. Currently, I get the following:
>
> under 2.125/2.126 SMP:
> machine locks up after compiling 75-80 series. Sometime messages about being
> unable to free memory appear on the console.
>
> under 2.0.36-pre-laest SMP: kernel oopses after compiling 10+ series.
>
> under 2.0.36-pre-latest UP: no problem whatsover compiling 100 series.
>
>
> The problem is not likely to be hardware related as far as CPUs or memory are
> concerned. (The test on UP kernel secceeds, no messages "internal compiler
> error... signal 11" appear). My setup is:
>
> Dual Supermicro P6DGU 440GX
> Dual PII 333 MHZ (no overclocking) with Deschutes core
> Onboard Adaptec 7890 Ultra II+
> SCSI Ultra II+ Quantum Atlas II QM39100AL-SCA 9.1GB, 7200RPM, 8ms, 1M cache.
> 256 MB PC100 RAM.
>
> I do not know what the problem is. I am going to investigate further, possibly
> contact the kernel guys.
>
>
I gave up on the 2.0 kernels and SMP. The 2.1's work flawlessly in this
regard. I'm running them here in production, with no hiccups of any kind
so far (knocks on desk superstitiously)
