Performance analysis

Hello,

We are having problem of performance on our server(Solaris 8).

I wish you guys could help me to analyse the result of various users commands to find out why and how to make our solaris feel better...

Thanks in advanced to all the people that will read and/or help me in this task.

1st, the load average

# uname -a

SunOS sivertbib6g 5.8 Generic_117350-34 sun4u sparc SUNW,Sun-Fire-V890

# top

Here is a short part of the "top" command i wrote on the shell

load averages: 3.14, 3.05, 3.1015:01:21

404 processes: 402 sleeping, 2 on cpu

CPU states: 40.5% idle, 16.7% user, 21.8% kernel, 20.9% iowait, 0.0% swap

Memory: 16G real, 4555M free, 9881M swap in use, 4714M swap free

I believe that when load average get over 1.X, we experiment lag.

So when it's 3.X, that's a very huge lag...

Could anyone from Sun help me confirm that theory about the limits of servers and load average? Or make any recommandation about it?

I also believe that 20% of iowait also doesnt help...

We use raid5, i heard that it is slow technology.

Is 20% of iowait a big amount ?

Could raid5 be the reason of such an amount ?

Could it be a broken/bad harddrive that makes the system goes slower?

What are the official solaris recommandation for the iowait ?

CPU seems to be ok, it has enough to compute, RAM memory seems too... But i may be wrong.

I'm a user (not root) so i cant do much stuff on that platform. But if you want me to run specific command to complete the post, i would do it.

Once again, thanks very much for you guys help.

[1696 byte] By [PierreHa] at [2007-11-26 20:55:08]
# 1

don't concentrate too much on load averages.

It reduces a complex system on just a few figures.

And on the figures you have I would't worry at all.

3.14 means just ~3 waiting threads (average) within the last minute.

Your server ships with at least 2 CPUs, so a load average with 2 could be normal. But to get a complete picture, you must go away from this figures and gather more data, eg feels the system slow or are you just worried about the 'top' result?

20% is not a big amount, 100% would be.

A broken or bad disk would be definitely logged into the syslog.

-- Nick

der_nikia at 2007-7-10 2:22:38 > top of Java-index,Sun Hardware,Servers - General Discussion...