Takeshi Nakayama <t...@catvmics.ne.jp> wrote:
> "sysctl -w vm.swapout=0" helps you?
It yields an interesting result: I now have a partial deadlock, with is
limited to the PUFFS filesystem. It is now possible to login and work in
the shell, something that was not possible with vm.swapout=1
However, even after memory is freed, the PUFFS filesystem does not
resume operation. The PUFFS userland server is perfused, which sends its
requests to glusterfsd. perfused is locked in memory thanks to a
mlockall(2) call, but glusterfsd can be paged out (it is probably too
big to get the mlockall(2) treatment). ddb shows me that:
242 4 3 0 4 c9d92060 glusterfsd anonget2
> 3 7 0 4 cacaf080 glusterfsd
2 3 0 84 cb0a0820 glusterfsd sigwait
1 3 0 4 cacaf300 glusterfsd anonget2
229 1 3 0 84 cacaf580 perfused kqueue
ddb's show uvpexp tells us we have a lot of free memory, and paging=114
does not change, which suggests this is only PUFFS-backed pages.
Current UVM status:
pagesize=4096 (0x1000), pagemask=0xfff, pageshift=12
63176 VM pages: 30432 active, 15057 inactive, 10183 wired, 2004 free
pages 34784 anon, 15634 file, 3874 exec
freemin=256, free-target=341, wired-max=21058
faults=1878379, traps=1970502, intrs=2721627, ctxswitch=3540426
softint=3556429, syscalls=-1829155228, swapins=0, swapouts=0
fault counts:
noram=0, noanon=0, pgwait=8, pgrele=0
ok relocks(total)=9608(9608), anget(retrys)=480235(7068),
amapcopy=299322
neighbor anon/obj pg=315480/3581619, gets(lock/unlock)=879980/2540
cases: anon=331217, anoncow=114469, obj=727628, prcopy=152352,
przero=545991
daemon and swap counts:
woke=33, revs=32, scans=45588, obscans=1244, anscans=32707
busy=0, freed=33837, reactivate=5243, deactivate=60890
pageouts=4110, pending=28677, nswget=7064
nswapdev=2, swpgavail=98303
swpages=98303, swpginuse=27376, swpgonly=20320, paging=114
--
Emmanuel Dreyfus
http://hcpnet.free.fr/pubz
m...@netbsd.org
