CPU utilization breakdown

[Jeeshan]

Hi All

please take a look at my scenario first,

previously:
model:7420, cell:1,processor:4 (dual core) PA-RISC, memory:16GB,OS:11iv1

current (after cell,memory,CPU upgradation):
model:7420, cell:2,processor:8 (dual core) PA-RISC, memory:32GB,OS:11iv1

my question is, if there is any possiblity to get more system resources by OS itself after upgrading the previous scenario?

and if there is any command to view the system processes itself utilizing CPU(s)? i need it in aggregatedly. or if there is any tools to check CPU utilization breakdown like 'meminfo'?

[Rasheed Tamton]

Hi Ahsan,

If I get your questions clearly:

>if there is any possiblity to get more system resources by OS itself after upgrading the previous scenario?
You have to fine tune the related kernel parameters to fully utilize.

>and if there is any command to view the system processes itself utilizing CPU(s)? i need it in aggregatedly. or if there is any tools to check CPU utilization breakdown like 'meminfo'?
glance -a (CPU By Processor) avg
top will give you individual CPU aggregate
sar -uM (mulitprocessor) or -p (pset)
UNIX95= ps -e -o "user,pcpu,cpu,vsz,pid,ppid,args" | sort -nk2

Are you looking for the individual system processes.

scopeux / measureware with extract (/var/opt/perf/parm) might give you some useful info.

HP might have some internal tools. Hope someone would post the info.

rgds.

[Jeeshan]

Salam Rasheed

>>You have to fine tune the related kernel parameters to fully utilize.

which kernel parameter? dbc_maxa_pct and dbc_min_pct are all set to 10 and 2 respectively. any other parameters need to modify?

>>Are you looking for the individual system processes.

no. just need to identify the summary how much CPU utilization concisely. but not with your commands or glance. any other tools or command i can find this?


i have only glance plus installed. no measureware installed.

sorry Rasheed, i'm not satisfied.

[Jeeshan]

i must apprfeciate some inputs from the experienced specialists.

;-)

[Pete Randall]

If you look at Glance's CPU processor reports (Glance > Reports CPU Info > CPU Report *OR* CPU By Processor), they will both show you the percentage of CPU consumed by System Calls - either in total (CPU Report) or by CPU (CPU By Processor).

If I interpret your question correctly, that should give you what you want.


Pete

[Rasheed Tamton]

Hi Ahsan,

Back from our weekend.

>which kernel parameter? dbc_maxa_pct and dbc_min_pct are all set to 10 and 2 respectively. any other parameters need to modify?

glance -m would show the memory report. Look at the footer. Look the system mem, Buf Cache and Free mem.

>model:7420, cell:2,processor:8 (dual core) PA-RISC, memory:32GB,OS:11iv1

You have 32GB mem, then I would recommend to change dbc_max_pct to 3 and dbc_min_pct 1 respectively. That would free more than 2GB mem. Also use swapinfo and vmstat.

Other parameters like shmmax, sem* and others depends on your application (32 bit or 64 bit) and database (oracle, etc). Also check the maxdsize, timeslice, etc.

sar -v 5 15 would give you the kernel table status. It is same as glance -t. You can fine tune all those kernel parameters (nproc, nfile, etc.) as per the usage.

>no. just need to identify the summary how much CPU utilization concisely. but not with your commands or glance. any other tools or command i can find this?

Performance means spending time. Sometimes, it is difficult to get responses on the forum because it demands spending time.

Unless you use some metrics, you can't really measure the system performance. If you do not have any specific reason to worry, just monitor the system for a while by collecting some metrics. There are no other tools as far as I know other than the ones I mentioned above, except HP uses some internal ones.

Regards.

[Rasheed Tamton]

The below link would give info reg. the Configurable Kernel Parameters.

http://docs.hp.com/en/939/KCParms/KCparams.OverviewAll.html

Regards,
Rasheed Tamton.

[Hein van den Heuvel · Accepted solution]

>> possiblity to get more system resources by OS

Why? The OS does what it needs to do to support the application.
Think APPLICATION.

Yes, you may be able to tweak the system to more efficiently support the application, but only after knowing what it does.

For example system MAY have had dbc_max/min set to 10/2 because any more would have put too much memory pressure on the box. Now you have more physical memory. Automatically the OS will have more dbc to play with because it is a percentage. But you now may have 'excess' memory and you may be able to improve the system by setting dbc_max/min to 20/5... but only if caching more file helps the application. If the application does not significantly use files then sticking to the actual quantity might be be better and you would have to reducd to 5/1... but that is more likely to hurt than help IMHO.

If you had configure a database with 4GB of buffer memory, then you may be able to increase that to 12GB and with that use more system resources.

So to help, you need to THINK APPLICATION
It defines how many resource are use, and can be used. To take 2 extreme example, if all you have is 1 monster program, the all you need is 1 1/2 CPUs and any more will not help.
If you had a backend machine with 8 service slaves on the older 4 cpu configuration, then it may be time to configure 16 service slaves.

What is the system being used for?
(Oracle) Database server?
File & Print? Development box? Front-end (users directly connected) or Back-end (behind transactional or application layer). a Few big programs or many little ones?

What prompted the system increase? Excess money? Probably not. So what measurements a nd arguments were use to convince folks to make an investment? Well, go back and re-run that evaluation.

To help us help you, you need to describe how the system is being used! HPUX is already doing what is best 'in general' you have to provide specifics and you gave none so far.

Other tools you may want to try:
- vmstat
- ipcs

hth,
Hein van den Heuvel
HvdH Performance Consulting.

[Jeeshan]

thanks all.