Well, what a week! I ran two copies of this useful script, one on each
physical disk , while simultaneously running BOINC clients at nice=19
to fill in the CPU gaps. Sometimes I would try an emerge -e world
instead of one of the memtest scripts; I did that because I wasn't sure
I'd cleaned up the tool-chain since installing, so I emerged
linux-headers, glibc, binutils and gcc and then -e world.
I still kept getting segmentation faults, and other compilation errors
occurred in a few packages such as xorg-x11. Each segmentation fault
disappeared on running emerge --resume, as expected. Eventually I
remembered revdep-rebuild; running that cured the xorg-x11 problem (I
forget which library was rebuilt). The memtest script passed every time
- and the disk temperatures rose anything up to 20 C to over 50 C, which
illustrates the amount of extra work they were doing.
So I have a box that passes all tests except one: compiling programs.
Looks like I'll just have to wait for a harder fault to appear. Thanks
to all who've offered suggestions.
Linux Counter 5290, Aug 93.
 I did notice one minor wrinkle. The two copies started in series, so
one was always following the other with its requests for disk transfers,
at least at first, and so fell steadily behind. But by specifying, say,
40 passes, I saw that the two instances eventually drifted well out of
sync; I deduce therefore that the occurrence of interrupts will have
been random for all practical purposes. Secondly, I used a kernel
without swap because I thought it would make no difference; the disks
were being stressed equally anyway. Thirdly, I ran some tests from
xterms, and others from the console without X running.
email@example.com mailing list