Kernel too big - what to tune.

[Stephanie Nicholls]

ive a itanium box running HPUX 11.23.0409 and Oracle 9i with 4Gb of physical RAM. According to glance sys mem is 1.0Gb. While its not excessively huge and we dont have any paging issues I dont see the point of wasting RAM on the kernel if its not necessary.

I downloaded the kmeminfo tool but cant see anything obvious that would make a big difference. Any Ideas ??

kmeminfo (5.01)
unix: /stand/vmunix 11.23 64bit IA64
core: /dev/kmem live
link: Sat Nov 27 14:40:37 EDT 2004
boot: Sat Aug 20 20:18:26 2005
dump: Tue Sep 6 15:13:09 2005
nbpg: 4096 bytes

----------------------------------------------------------------------
Physical memory usage summary (in page/byte/percent):

Physical memory = 1045698 4.0g 100%
Free memory = 74254 290.1m 7%
User processes = 596419 2.3g 57% details with -user
System = 362254 1.4g 35%
Kernel = 278599 1.1g 27% kernel text and data
Dynamic Arenas = 169656 662.7m 16% details with -arena
vx_inode_cache = 40128 156.8m 4%
vx_global_pool = 23719 92.7m 2%
M_TEMP = 22749 88.9m 2%
vx_buffer_cache = 16960 66.2m 2%
spinlock = 14549 56.8m 1%
Other arenas = 51551 201.4m 5% details with -arena
Super page pool = 24996 97.6m 2% details with -kas
Static Tables = 56164 219.4m 5% details with -static
pfdat = 24508 95.7m 2%
vhpt = 8192 32.0m 1%
nbuf = 8064 31.5m 1% bufcache headers
text = 6919 27.0m 1% vmunix text section
bufhash = 2048 8.0m 0% bufcache hash headers
Other tables = 6432 25.1m 1% details with -static
Buffer cache = 83655 326.8m 8% details with -bufcache

Thanks

Steph

[RAC_1]

What are your settings for
vx_ninode, buffer cahe??

[Stephanie Nicholls]

vx_ninode is set to the default which is 0

dbc_min_pct = 4, dbc_max_pct = 8

cheers

Steph

[RAC_1]

I think, posting kmtune -l or sysdef will help.

[Stephanie Nicholls]

Heres a copy of the sysdef (its a lot shorter than kmtune output)

NAME VALUE BOOT MIN-MAX UNITS FLAGS
acctresume 4 - -100-100 -
acctsuspend 2 - -100-100 -
allocate_fs_swapmap 0 - - -
bufpages 83655 - 0- Pages -
create_fastlinks 0 - - -
dbc_max_pct 8 - - -
dbc_min_pct 4 - - -
default_disk_ir 0 - - -
dskless_node 0 - 0-1 -
eisa_io_estimate 768 - - -
file_pad 10 - 0- -
fs_async 0 - 0-1 -
hpux_aes_override 0 - - -
maxdsiz 262144 - 0-655360 Pages -
maxdsiz_64bit 1048576 - 256-1048576 Pages -
maxfiles 2048 - 30-2048 -
maxfiles_lim 4096 - 30-2048 -
maxssiz 2048 - 0-655360 Pages -
maxssiz_64bit 262144 - 256-1048576 Pages -
maxswapchunks 0 - 1-16384 -
maxtsiz 24576 - 0-655360 Pages -
maxtsiz_64bit 262144 - 256-1048576 Pages -
maxuprc 3200 - 3- -
maxvgs 10 - - -
msgmap -536870911 - 3- -
nbuf 46872 - 0- -
nfile 65536 - 14- -
ninode 4880 - 14- -
no_lvm_disks 0 - - -
nproc 4200 - 10- -
npty 60 - 1- -
nstrpty 60 - - -
nswapdev 10 - 1-25 -
nswapfs 10 - 1-25 -
public_shlibs 1 - - -
remote_nfs_swap 0 - - -
rtsched_numpri 32 - - -
sema -536870911 - 0-1 -
shmem -536870911 - 0-1 -
shmmni 512 - 3-1024 -
streampipes 0 - 0- -
swapmem_on 1 - - -
swchunk 2048 - 2048-16384 kBytes -
timeslice 10 - -1-2147483648 Ticks -
unlockable_mem 0 - 0- Pages -

[RAC_1]

As the values of nbuf and bufpages is non-zero, you are using static cache. What is the output og glance -m

What is the buffer cache reported on glance -m.

Also the values
nfile 65536
ninode 4880

Look bit high. Most proablly, /stand will be only hfs (to which ninode setting applies)
file system. You can bring it further down. (to 2096 or so)
nfile looks too high. This is system wide open files parameter. Why it has been set that high?? On my system (running oracle and apache web server) it is set to 30000.
Check it 's real time usage as follows.
sar -v 2 10
glance -t

If it is not hitting beyond say abuot 30000, then lower that setting.