problem of swap

[Michael Steele_2]

Your right V. Nyga. But I also got to say, I don't know anyone who goes to swapinfo to see how much phyical memory is being used. Everyone goes to glance. Its very readable and very accurate. And very free.

[Dennis Handly]

>Michael: Everyone goes to glance. It's very readable and very accurate. And very free.

glance isn't free. There is a trial version available.

[kacou]

How can i execute "glance" on my server hp-ux?

# uname -r
B.11.23
# model
9000/800/rp3440
#

[Grayh]

This should be of good help to you:-

HP-UXdev # glance

B3692A GlancePlus C.03.71.00 18:21:16 nedunix2 9000/800 Current Avg High

--------------------------------------------------------------------------------

CPU Util S SN NU | 43% 59% 95%

Disk Util F F | 33% 263 38%

Mem Util S SU | 87% 87% 87%

Swap Util U UR | 81% 81% 81%

--------------------------------------------------------------------------------

CPU REPORT

-------------------------------------------------------------------------

State Current Average High Time Cum Time

-------------------------------------------------------------------------

User 6.3 4.3 7.0 0.29 1.18

Nice 10.7 3.8 10.7 0.49 1.05

Negative Nice 0.2 0.3 1.3 0.01 0.08

RealTime 0.2 0.2 0.2 0.01 0.05

System 5.7 6.3 9.0 0.26 1.73

Interrupt 0.7 1.1 1.8 0.03 0.30

ContextSwitch 0.0 0.3 0.7 0.00 0.07

Traps 0.0 0.0 0.0 0.00 0.00

Vfaults 1.3 0.7 1.3 0.06 0.18

Idle 74.8 83.0 91.3 3.44 22.68


Top CPU user: PID 8143, perl 7.8% cpu util

Active CPUs: 2

[Dennis Handly]

>How can I execute "glance" on my server hp-ux?

Which? You mention 11.31 above and now 11.23.

B3701AA C.04.70.000 HP GlancePlus/UX Pak for 11.31
B3693AA C.03.86.00 HP GlancePlus/UX for s800 11.23

The trial ones may be close these numbers. It should be on your OS installation CDs.

Once it is installed:
/opt/perf/bin/gpm # GUI version
/opt/perf/bin/glance

[Michael Steele_2]

Gee Wiz Dennis. The trial version is free. It runs for awhile and then stops and asks for a license.

In this case the author can download the trial version right here:

https://h10078.www1.hp.com/cda/hpms/display/main/hpms_content.jsp?zn=bto&cp=1-11-15-28^9637_4000_100__

[Emil Velez]

There are a few parameters.. You do not specify the version of HPUX so some parameters might be obsolete.

maxswapchunks may be set to 64000
swchunk normally set at 2000

also look at dbc_min and dbc_max see if dbc_max is set to 50 lower that

it looks like you have enough swap space but it cannot reserve any more.

also look at your nproc kernel parameters

may want to post the values of these and the HPUX version

[Don Morris_1]

How do you add more swap? Reduce your virtual load (by throttling down or stopping jobs) such that you have sufficient free swap to let swapon run (you show that you hit 100% swap consumption just trying to get it to run originally).

Certainly if this is your normal workload -- you'd be better off adding at least 3.3Gb (round up to 4Gb) of physical memory. Your first swapinfo shows that much paged out to your swap device... and hence, you're killing your performance waiting for swap to go in and out. Alternately, decrease your workload to conform to your hardware.

Since this is 11.31, what's your filecache_min and filecache_max? (You don't want filecache_min too high... that would reserve a lot of RAM for file cache only).

[kacou]

HP 9000 rp3440 Server
# model
9000/800/rp3440
# uname -r
B.11.23
#

[Don Morris_1]

Fine.. then on your 11.23, you do want to worry about dbc. (You also posted:

glance doesn't work on my server. What can i use another?

# uname -r
B.11.31
# model
ia64 hp server rx7640
#

upthread, hence my pointing out that the dbc max/min tunables now roughly correspond to the filecache ones.

The root is still the same. You couldn't add swap at the original message because you were too exhausted to run new user programs (like swapon itself). So you have to free swap up first by decreasing the load. How you do that is your call. The output there also showed over 3Gb paged out, which strongly indicates you don't have enough RAM to run what you are trying to do anyway, so you have to either add more memory to the system or (again) decrease the load.