Question : UNIX Performance Monitoring

Sometimes, some users of our system report that the system is very slow.  Is there any procedure that can help us the Unix OS status e.g. swap memory etc.

Also, is there any web resources that teach us about the system administration of a UNIX machine.

Answer : UNIX Performance Monitoring

There are several commands you can use with HP-UX 10.x to help you troubleshoot a performance problem.  The system can appear to be slow when there is one of 4 possible bottlenecks: CPU, memory, disk or network.  Provided below are some usefull commands to help you determine the bottleneck:

# sar -u 2 100
Will tell you how much time during the interval the CPU is spending in user, kernel and waiting for I/O modes.
Look for values > 80%

# sar -v 2 100
Will show you how your kernel tables are doing

# sar -b 2 100
Will show you how you are using your buffer cache. If you see low %rcache and %wcache values (these are your hit rates for buffer cache) then you are not using your buffer cache efficiently. 80% to 100% is optimal.

# sar -d 2 100
Will show you perfomance metrics on each device.  avserv should be lower that avwait and avqueue should not be higher that 1-10.  

# vmstat 2 100
Will show you your virtual memory metrics. sr should be very low (0-5) po should also be low (0-5). pi (page ins) are OK, it is the pageouts you need to be concerned about

# swapinfo -tma
Will show you how your devices are being used for swap activity. Want to see low values for PCT USED field (0-10).  PCT USED values coorespond to how swap is used for pageout activity.

# iostat 2 100
Will show you seek times and transfer rates of your devices.

# netstat -i
Will show you network packet traffic and collision rates.

Also GlancePlus is a must for troubleshooting performance problems. You can try the trial version on the Application CD for free.

Feel free to post some of the output of these commands if you are not sure how to interpret their meaning.

Tony