Load average

[John Jayaseelan]

Hi,

The top command shows values (1.70, 2.32, and 2.64) on the Load average line. Could anyone please tell what does it mean and how that can be interpreted.


Does it mean that the system is just loaded with 1.70%, 2.32%, and 2.64% of its capacity?


Thanks
John Jayaseelan

[Elmar P. Kolkman]

See man uptime(1).

The numbers are the average number of jobs in the run queue over the last 1, 5 and 15 minutes. So you have a system that is getting quietter over the last 15 minutes...

Hope this answers your question.

[Graham Cameron_1]

The 3 numbers are load averages in the last one, five, and fifteen minutes of all the active processors in the system.

Load average is absolutely not the same as cpu usage. It's a weighted figure to give an indication of how loaded the system is, based on cpu usage, run queue, and other stuff.

A healthy load average is generally in the low single figures. If it's > 10 you're in trouble.

There are loads of threads in the forums, here's one but if you search you'll find many more.

http://forums1.itrc.hp.com/service/forums/parseCurl.do?CURL=%2Fcm%2FQuestionAnswer%2F1%2C%2C0x6cd29200caded5118ff40090279cd0f9%2C00.html&admit=716493758+1066736474212+28353475

-- Graham

[Michael Steele_2]

Disregard top when it comes to performance evaluation.

Load average is the number of jobs in the CPU run queue. With multiple CPU's you devide by the number of CPU's.

An old definition from O'Reily Performance Tuning which was applied to all Unix platforms and not confined to HP-UX, defined a bad load average when the run queue was greater than 1. I believe this is obsolete.

A CPU bottleneck defined in HP MeasureWare is a combination of three metrics, high load average (* same definition as O'Reily *), Higher than usual number of processes running plus 100% CPU utilization.

This is the metric I rely upon for a CPU bottleneck.

That, and %busy = 99% in 'sar -u'.

[John Jayaseelan]

Thanks to all for the Info.

What is 'HP MeasureWare', is it a tool like top?

The O/S is HP-UX 11.00, sar command is not accessible.

Thanks

[Graham Cameron_1]

sar should be in /usr/bin/sar on hp-ux 11.

[Michael Steele_2]

Regarding "..sar not accessible.."

Login as root.

# /usr/bin/sar -u 5 5

Also refer to glance or gpm. A free version will be on your diagnostics cd's.

[John Jayaseelan]

Hi,

The following is reply for the sar.

# type sar
sar is /usr/sbin/sar
[cctest01]: /root
# sar
sar: Can't open /var/adm/sa/sa21

Michael,

You mentioned about 'HP measureview', What is this measure how this can be evaluated using which O/S tool?

Thanks

[Elmar P. Kolkman]

If your sar files are not available, you will have to start your sar collector process to create them. This is normally done using cron to start it regularly, for instance:
0 * * * * /usr/lbin/sa/sa1 300 12

As for measureware: for this you need extra codewords. To see if it runs, do a 'ps -ef | grep scopeux'. Or 'swlist | grep -i measure'.

[T G Manikandan]

It should be like
#sar t n

t-->time interval
n-->number of times
#sar 5 5

5 times in 5 sec interval

[Michael Steele_2]

HP MeasureWare will be on the same cd as the glace cd. You may need a codeword, you certainly need the dictionaries. Oftentimes these dictionaries come with the server and will be with the server documentation.

There are many phone numbers to HP if you need a codeword: 1800-386-1115, 1800-633-3600.

I think you ought to load some more recent diagnostics. A.42 is the current version.

Sar should be invoked through crontab for root.
0 * * * * /usr/lbin/sa sa1 1200 3
Should collect data at 20 min. interval
55 23 * * * /usr/lbin/sa/sa2 -s 0:00 -e 23:55 -i 1200 -A

[Bill Hassell]

The value for top's load average is identical to the uptime command. It is the avergare size of the runqueue at 1,5,15 minute sample sizes. It is a very crude indication of load, but an accurate representation of the runqueue. What is the runqueue? It is the number of processes that are either running right now or ready to run but no CPU is available. On a single CPU system, a runqueue of 1 means that one process is using the available CPU, but qa runqueue of 2 means one is running and the other is waiting for a timeslice. Remember that these are LONG averages (minutes) and the opsystem will switch between processes as much aas 10 times per second.

So how does this relate to a workload? Only indirectly. You may have a system that runs 500 processes at the same time and each process runs for 5ms then sleeps for 500ms. The runqueue could be as high as 20 or 30 (30 processes waiting to run on average) yet the system seems to respond very fast. Tat's because the queue is processed very rapidly within human terms (seconds). In another situation, the runqueue might be 4 and the system seem quite slow because there is only one or two processors. In the first example, system overhead would be quite high (up to 60%) but in the second case, only more processors will improve the performance.

So use uptime and top as crude pictures. If you really need to monotor the system, load Glance from your application CDROMs. It will run for a month or two before you have to purchase the licenses for it. Measureware is the long term data collection package whereas Glance gives you short duration snapshots of the system.