Unix performance

"top" for load average & CPU hogs
"sar" for system activity, e.g. cpu, disk activity, swapping

How to tune it is going to be dependent on your applications, but general tips are:
If swapping is occurring, performance will start to suffer - Make sure your swap partitions are on fast disks, or get more memory.
Balance the load on your disks

What should I do to balance the load and using sar -u option I found that the statistic would look like this
%usr %sys %wio %idle
20         13     37      30    

is it the inefficent of the I/O sub system and what to be tune for what further step to use.

..

My server have the program using share memory, message queue, and semophore and how to check each process take how many memory and what are the kermal parameter I should put to make sure it will not exist shmmax parameter?

On first,  your question is very hard. and wich UNIX you use ?


This is very complex process and you must work
long monitornig to determine your problems.

You need answer me on several question.

Which UNIX you use?
Do you use some database systems ?
Please  explain all aspects of your system and describe (.
hardware, software, application etc...)
 

1. Tune your aplications (this is 60 % of tuning any system) if application
working slow, any tuning od operating system can't help.

2. Tune your memory (paging, swaping, kernel )

3. Tune your disk access and balance your physical organization of data.
 

%usr %sys %wio %idle
  20         13     37      30

This is only one moment and  this is not enought for any answer.  

Kernel tuning is not all tuning of one system.


Send my list of all your kernel parameters.

Sorry but on my opinion you don't undrstend tuning.

sar -u only shows CPU utilisation; At the moment you took that snapshot, your system was lightly loaded (30% idle) and the CPU was waiting for data to be read from disk (swap or filesystem), which takes thousands of times longer than to read from cache (memory); This does not imply inefficiency: If the system was more heavily loaded, it would be doing other stuff while waiting for the data.

"top" will show you CPU & (virtual) memory usage of your processes (If you don't have top, ps can usually show this - Which Unix do you use?)

You need to look at the sar averages over a few hours; CPU utilisation in itself does not give you the full picture. You need to look at whether it is swapping processes from memory to disk (slow) and how much i/o there is on each disk (very slow)

To expand on globalv's expansion on my previous generic tips:
Application tuning: Put O/S, swap areas, applications and data on seperate disks to spread the load. You will need to experiment with how to spread the data out, but the application provider can give you tips, e.g. for Oracle, put /archive (redo logs) on a seperate disk from the Db tables

There will be a kernel parameter to specify max % of memory that can be used for caching data. If you have lots of free memory (sar), increase this maximum to e.g. 50% (Note, if other processes need the memory, it will be released to them).


Performance tuning is a BIG subject; An expert would need several hours monitoring and analysing your system and it's workload to tune it.

The MINIMUM I would want to know would be:
Hardware details: Memory, SCSI layouts, disk info. and filesystems;
Details of your applications and the associated data transfers;
Sar averages (all statistics) over the period you feel the system is "slow";

What is goal of this tuning ?