(Sorry if the quoting got FUBAR.)
I tried this "once" more, with 1GB VM RAM, no ARC settings and a 4GB
swap for dumps. (Apparently, when I had 640MB VM RAM, the dump created
was ~1150 MB. I figured it couldn't exceed RAM size.)
Between the previous tests and this, I had loads of crashes (with
640MB), even so bad that I couldn't boot because savecore would cause
a panic. I increased VM RAM and set arc_max in the loader and it
booted fine. Then, I tried (see below) with 1GB VM RAM and again no
ARC settings.
The wired count grew and grew and grew, until it crashed in
lzjb_decompress(), backtrace:
lzjb_decompress()
zio_decompress()
zio_done()
zio_execute()
zio_done()
zio_execute()
taskq_thread()
fork_exit()
fork_trampoline()
On "call doadump" I got "Fatal double fault", no dump, and a reboot.
Here's a LONG output of some vmstat output while running buildworld -
j4. Note how the wired count keeps increasing and increasing until it
breaks (in part, I guess this is intended, but it seems to grow a tad
out of hand):
Regards,
Thomas
------ NO arc settings below! ------
[serenity@clone ~]$ while :; do date; echo; vmstat -s | grep -E 'pages
(cached|active|wired down|free$)'; echo; sleep 20; done
Wed May 27 13:54:50 CEST 2009
15716 pages cached
41358 pages active
63244 pages wired down
141368 pages free
Wed May 27 13:55:10 CEST 2009
15719 pages cached
65571 pages active
65515 pages wired down
113358 pages free
Wed May 27 13:55:30 CEST 2009
15719 pages cached
41471 pages active
70436 pages wired down
132093 pages free
Wed May 27 13:55:50 CEST 2009
15809 pages cached
11660 pages active
81518 pages wired down
153294 pages free
Wed May 27 13:56:10 CEST 2009
16124 pages cached
11013 pages active
87862 pages wired down
147672 pages free
Wed May 27 13:56:30 CEST 2009
16124 pages cached
10917 pages active
...
