need to check kernel tunables

[Vipin Singhal]

My server seems to have 16 GB of memory, but the currently run program (and system) seem to require about 18 GB of memory. This causes pages thrashing to and from the virtual memory.
I need to check kernel tunables and physical memory on server.

how i will do this?
thanks
vipin

[Thummalu]

Hi Vipin,
pleaee use "kmtune" to know about your kernel parameters.

or as root user use SAM to know about them.

to check your memory status,
run the following command:

swapinfo -tam

Br
Thummalu

[john kingsley]

vmstat & top will show you this information.

Check dbc_max_pct & dbc_min_pct (kmtune | grep -i dbc)
By default, HPUX sets these values at 30% which is way too high. The lower you make them, the more physical memory will be available. Unless your application specifically requires a higher value, you could probable set them to 10% or less.

[Pete Randall]

The first thing to check is dbc_max/min. It defaults to 50% and that is far too high. Typically somewhere around 300 to 500 MB is more than enough for buffer cache so you could tune this down to 4% or 5% and free up some memory.


Pete

[Geoff Wild]

What is the currently run program? Oracle? - You may need the dba's to tune the db so it uses less memory - IE - decrease size of sga...

execute kmtune to see your current parameters...

You want the dbc something like the following or less:

dbc_max_pct 5 - 5
dbc_min_pct 4 - 4


Rgds...Geoff

[Vipin Singhal]

#kmtune | grep dbc
dbc_max_pct (now) 30 30 Immed
(next boot) 8 8
dbc_min_pct 5 Default Immed

# swapinfo -tam
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 4096 121 3975 3% 0 - 1 /dev/vg00/lvol2
dev 20480 127 20353 1% 0 - 1 /dev/vg00/swap2
reserve - 342 -342
memory 16351 5073 11278 31%
total 40927 5663 35264 14% - 0 -


how much memory i am using ?
is all the things are all rights?
does it need to make some changes?
thanks
vipin

[Geoff Wild]

With 16GB of ram - you should drop dbc_max_pct to 5

Memory? swapuinfo reports on device and file system paging space.

The memory line is the memory paging area (also known as pseudo-swap).

Not the amount of memory your system is using...

A good utility is the memdetail program - I attached the source code - just compile it with cc

Output like:

# memdetail
Memory Stat total used avail %used
physical 14240.0 13510.4 729.6 95%
active virtual 13650.7 6720.8 6929.9 49%
active real 10650.0 5058.9 5591.1 48%
memory swap 11139.4 1838.7 9300.7 17%
device swap 26528.0 13031.9 13496.1 49%

Rgds...Geoff

[James R. Ferguson]

Hi Vipin:

Your best friend for performance monitoring is 'glance'. There is built-in help that gives a general definiation of each metric too. If you don't already license it, you should.

Regards!

...JRF...

[Vipin Singhal]

I have glance:
this is the output of glance:

GlancePlus C.03.85.00 10:18:09 warbler ia64 Current Avg High
-----------------------------------------------------------------------------------------------------------------------------
CPU Util | 0% 0% 0%
Disk Util F | 1% 1% 2%
Mem Util S SUUB B | 58% 58% 58%
Swap Util U UR | 14% 14% 14%
-----------------------------------------------------------------------------------------------------------------------------
PROCESS LIST Users= 9
User CPU Util Cum Disk Thd
Process Name PID PPID Pri Name ( 800% max) CPU IO Rate RSS Cnt
--------------------------------------------------------------------------------
vxfsd 57 0 134 root 0.2/ 4.4 98562.8 1.3/ 1.8 3.6mb 47
midaemon 2015 1 -16 root 0.0/ 0.2 4381.7 0.0/ 0.0 38.2mb 2

[Patrick Wallek · Accepted solution]

At the time that glance was run, you were using 58% of your RAM. That's pretty good.

It appears that at some point you may have paged out though.

If you start noticing that things slow down, bring glance up and check the memory usage. If it hits 100% for prolonged periods, then you may have a valid case to upgrade the RAM in the machine.

[James R. Ferguson]

Hi (again) Vipin:

You can see more details about overall memory utilization in 'glance' by toggling the (m)emory metric screen.

As Patrick noted, page-outs are a sign of memory pressure. You can monitor those here.

Regards!

...JRF...

[Geoff Wild]

You can also use vmstat to check paging:

vmstat -S 2 20

watch the po

man vmstat for more info...

-S Report the number of processes swapped in and out (si
and so) instead of page reclaims and address
translation faults (re and at).


Rgds...Geoff