Stuart Henderson wrote: > I've had faulty hardware that was somewhat stable with earlier > releases but crashed more often with code from sometime in april; the > key point is that the hardware _was_ faulty. memtest86 did not find > any RAM errors. 'make build' whilst running stress (from ports) did > crash (quickly in the case of the newer OS; after running for a while > in the case of the older OS). This was resolved by replacing the CPU > with a good one.
I ran a 'make build' Friday night on 4.0, along with 'stress --vm 1 --cpu 33' but it was still trying to configure apache after several hours, so I cancelled the stress, changed hw.setperf to 100 and let it grind away. The next morning I found the system frozen at the libc build stage (qsort.o). I'm not sure that any of this is really significant since under 4.0 this machine freezes up even when the system is completely idle for hours on end. Yesterday I reverted back to 3.9 and did another 'make build', with hw.setperf also at 100. It completed without any problems (no sig11, crash, freeze or anything out of the ordinary). If you think it's worthwhile to try that again with more load, what stress parameters do you suggest?
