(see original posting
http://www.ornl.gov/its/archives/mailing-lists/tru64-unix-managers/2001/09/msg00343.html
)
Hello !
Almost turns out to be an unclear issue of what the "Real"
line in the 'top' output means, although there are still
inconsistencies.
Nick Hill, John P. Speno, and Stiffler Durden
replied (thank you all). The former two are confident that
I'm essentially getting what I asked the machine to do,
except Stiffler, who suspects that I'm scratching some
limit somewhere. Nick was baffled about the 20G the kernel
idle process uses. I think this is the usual behaviour,
as the kernel grabs all memory for management (although
the amount does also not fit??). But I see a inconsistency
in the vmstat output: According to the Systems Conf and
Tuning Manual the sum of act,free,wired should be close
to the physical limit (16 GB), which does not hold:
========================= vmstat 2 ==================================
Virtual Memory Statistics: (pagesize = 8192)
procs memory pages intr cpu
r w u act free wire fault cow zero react pin pout in sy cs us sy id
6 242 171 1M 1604 98K 195M 4M 160M 37M 19M 2M 57 4K 326 45 0 55
6 242 171 1M 1683 98K 269 281 261 0 592 19 253 114 1K 25 1 74
^^^^^^^^^^^^^^^
========================= vmstat -P ==================================
Managed Pages Break Down:
free pages = 1854
active pages = 583738
inactive pages = 1167457 <===
wired pages = 98873
ubc pages = 205681
==================
Total = 2057603
It shows only 8 GB(1M) in use and 9GB inactive, and top shows that 10
of the 16 GB are presumably not in use, instead 12 GB of swap space is
filled:
=========================== top ====================================
load averages: 1.03, 0.99, 0.85 11:36:52
137 processes: 3 running, 65 waiting, 26 sleeping, 42 idle, 1 stopped
Memory: Real: 6167M/16G act/tot Virtual: 12396M/22954M use/tot Free: 15M
#########
PID USERNAME PRI NICE SIZE RES STATE TIME CPU COMMAND
4037442 xxxx 42 19 6283M 2168M run 39:06 99.90% Rcp_0.6.22 <===
4037615 xxxx 42 0 2616K 475K sleep 0:00 4.00% tcsh
3670016 root 0 0 20G 593M run 165:37 1.40% kernel idle !!!
4037430 xxxx 42 19 9797M 5341M sleep 38:53 1.30% Rcp_0.6.22 <===
So I wondered why top showed only 6G of 16G to be actively in use
(the "Real:" line), but the system at the same time had only 15M left.
Nick also stated that the missing memory may be used by something else,
but an extensive search (ps -e -o pid -ovsz -orss -ocommand, ipcs -a,
swapon -s, ulimit) did not reveal anything. This would also not fit into
the observation that when I only start one large process (~12 GB) all
outputs as expected, and the system does not page at all.
The processes itself are F90 programs using the parallel CXML libraries.
So I just suspect that the 10G difference in the "Real:" line is the
memory marked as inactive pages + wired/ubc, therefore closing this
case as a misunderstanding of the top output.
--
Dr. Udo Grabowski email: udo.grabowski_at_imk.fzk.de
Institut f. Meteorologie und Klimaforschung II, Forschungszentrum Karslruhe
Postfach 3640, D-76021 Karlsruhe, Germany Tel: (+49) 7247 82-6026
http://www.fzk.de/imk/imk2/ame/grabowski/ Fax: " -6141
Received on Fri Sep 28 2001 - 09:04:03 NZST