how to interprete uptime results ?

[Jeroen_D]

Hi,

I've checked the man uptime but did not find what to think of loads like 3.5

What is high and what is low ? What is acceptable ? At what point should I consider upgrading the HP-UX station ?
I am having a lot of 3.5 loads on the machine that is running an Oracle database. Should I install it on a more powerful machine ?

TIA,
Jeroen.

[Pal Szabo_1]

Hi!

The uptime shows, how busy is you cpu.How long wait the processes in the waiting queue ...

Regards:
Paul

[Dan Hetzel]

Hi Jeroen,

A high value for the run queue doesn't always mean that there is a problem. You could have a large numer of fast polling processes for example.
What are the CPU usage figures?
What are the values of dbc_min_pct and dbc_max_pct kernel parameters ?

dan

[Paula J Frazer-Campbell]

Hi
The load level depends on many things within your system processors, memory disks configuration etc.
So before jumping to a new/bigger machine search the forums on monitoring and high load problems.

HTH

Paula

[Jeroen_D]

I've instructed cron to run uptime every 10 minutes :

9:40am up 37 days, 44 mins, 6 users, load average: 3.52, 3.42, 3.31
9:50am up 37 days, 54 mins, 6 users, load average: 3.56, 3.44, 3.36
10:00am up 37 days, 1:04, 7 users, load average: 3.90, 3.57, 3.46
10:10am up 37 days, 1:14, 13 users, load average: 4.41, 3.82, 3.58
10:20am up 37 days, 1:24, 14 users, load average: 3.30, 3.52, 3.59
10:30am up 37 days, 1:34, 14 users, load average: 3.25, 3.36, 3.49
10:40am up 37 days, 1:44, 13 users, load average: 4.52, 4.29, 3.88
10:50am up 37 days, 1:54, 13 users, load average: 4.30, 4.30, 4.08
11:00am up 37 days, 2:04, 13 users, load average: 4.28, 4.35, 4.20
11:10am up 37 days, 2:14, 13 users, load average: 4.10, 4.23, 4.22
11:20am up 37 days, 2:24, 6 users, load average: 4.24, 4.20, 4.20
11:30am up 37 days, 2:34, 6 users, load average: 4.55, 4.30, 4.25

Some of top's highest processes :

TTY PID USERNAME PRI NI SIZE RES STATE TIME %WCPU %CPU COMMAND
p8 27940 root 234 20 108K 48K run 1162:54 22.60 22.56 telnet
p6 28039 root 233 20 108K 48K run 1239:08 22.18 22.14 telnet
? 439 blfv 234 20 128K 164K run 19:17 22.07 22.03 more
? 262 oracle 236 20 9412K 396K run 0:15 20.99 20.96 oracleSVP1
? 112 oracle 154 20 9524K 488K sleep 1:25 2.73 2.73 oracleSVP1
? 662 blfv 154 20 452K 916K sleep 0:02 0.57 0.56 dtterm
? 2 root 128 20 0K 0K sleep 57:38 0.49 0.49 vhand
? 482 oracle 154 20 9496K 608K sleep 1:01 0.35 0.35 oracleSVP1

What does a load of 3.5 mean ?

Jeroen.

[Jeroen_D]

FYI :

dbc_min_pct = 5
dbc_max_pct = 50

I've always thought that a load higher than 1 was a high load... correct ?

[Sandor Horvath_2]

Hi !

The uptime 3.5 mean average 3,5 process wait in run queue. 1, 5, 15 minutes average .

regards, Saa

[Dan Hetzel · Accepted solution]

Hi Jeroen,

A run queue of 3.5 means that there are in average 3.5 processes waiting to be "served".
But "system load" isn't a good indication of performance.

For systems running Oracle databases, it's quite generally accepted that dbc_max_pct shouldn't be left to 50. This could use too much memory and increase swap usage.

What is 'swapinfo -t' reporting ?

Depending on the amount of RAM you have, you could bring dbc_max_pct down to 20 and see if there is any improvement. You'll have to make a new kernel and reboot for changes to take effect.


Best regards,

Dan

[Jeroen_D]

I'll have to ask our Oracle specialist to tune the system a bit..

server-2:/ # swapinfo -t
Kb Kb Kb PCT START/ Kb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 262144 60200 201944 23% 0 - 1 /dev/vg00/lvol2
reserve - 93796 -93796
memory 89348 49912 39436 56%
total 351492 203908 147584 58% - 0 -
server-2:/ # dmesg | grep Physi
Physical: 131072 Kbytes, lockable: 87576 Kbytes, available: 103516 Kbytes
server-2:/ #

[Lasse Knudsen]

Loadavg shows only a little of the truth as far as CPU load is concerned. As one said it is the number of processes in the run queue. A process can be in the run queue when it is waiting for I/O but it does not necessarily mean that the cpu is loaded.

You should also use other tools like 'sar' to monitor your system. (try sar -u 10 10) here you will also see the amount of time the CPU is spending in waiting for I/O. Oracle has several processes when it is running and if all of them are waiting for I/O it can cause the high workload you see.

Monitor disk I/O also with sar (-d option).

[Dan Hetzel]

Hi Jeroen,

As your system only has 128Mb RAM, I would set the dbc_max_pct kernel param to something like 20-25.

This would free about 32Mb memory and possibly reduce swapping.

What do the other forum users think about this?

Cheers,

Dan