Re: unannounced p10.gz fixes lockups

ADAM Sulmicki (adam@cfar.umd.edu)
Sun, 21 Sep 1997 16:07:03 -0400 (EDT)


->Bonnie is a disk and filesystem benchmark. I tend to run either 5 with a
->200mb 'size' or 10 with a 100mb 'size'. You can find it in
->/pub/Linux/system/benchmark/bonnie.tar.gz on sunsite.unc.edu.

Cool thanks for info, but it looks it is not what I exactly look.

The way I prefer to stess kernel is about 33% I/O, 33% processing in
kernel (preferably not IO related), and 33% in user land.

the first I usually get by 'dd if=/dev/hda5 of=/dev/null' hda5 is in my
case 2 gb so it is usually enough (you can use 'hda' instead)

the 2nd I usaully get by 'ping -f -s 1024 foo'. Note a) the '1024' is
emprical value which is big enough to cause high load (kernel space), yet
small enough not to make kernel lose packets ..Also I have a local network
of two computesr so I prefer ping the machine being tested from the
different machine (so that ping does not use cpu) and the network card is
in use, and interrupts .. it can give you sometimes very interesting
results.

the 3rd, userland ... first i start netscape .. 4.03-pro preferably ..
(still I need good idea for a busy web site w/ java and animated gifs )
www.absnews.com seems be good choice ..

then I usaully start 'make -j zImage' for 2.0.31-pre10, but that is enough
I also have start, 'make World' for Xfree86-3.3.1, and 'make -j' for
sss-1.2.20

anything less than that will not kill kernel (even one make less). the
result of above in mine case is usually that loads beings to skyrocket,
plenty zobmies, and later differrent 'bad' things. [note it worked for me
w/ pre-9 .. I still have to see what happens w/ pre-10]

->What i've found to be a good stress test:
->ping -f localhost
->make -j zImage
->bonnie * 5
->netscape communicator
->emacs running gnus
->
->all at the same time. system is slow but still basically usable.
->(112MB RAM, Intel P133)
->

-Adam