Performance Tuning for Content Manager

284 Performance Tuning for Content Manager

11.3.1 Monitoring tools for AIX

For the AIX operating system, various monitoring utilities and tools are available

that provide information about internals of the operating system kernel, I/O

devices, and processes running. Although individual tools differ significantly in

the specific information they provide and the way they provide the information, all

of these tools internally interact to the same interface that the AIX operating

system provides. Before going into detail about any monitoring tool, we give an

overview of the measurement counters that AIX provides and that, from our

experience, have shown to be the most important for tuning a Content Manager

environment.

Measurement counters

The measurement counters that AIX and other UNIX systems provide for system

monitoring are commonly grouped into the following categories:

 CPU statistics

For each individual CPU, percentage of time that the CPU is in one of the four

states:

– SYSTEM: executing in system mode; that is, executing some application

in “kernel protected” mode or system call.

– USER: executing in user mode; that is, executing some application in

“unprotected” mode.

– WAIT: for example, processor is waiting for some disk or network I/O to

complete.

–IDLE

 Memory usage statistics

Several counters are performance related:

– Total size of physical memory, and the amount of physical memory used

– Total size of paging space, and the amount of paging space used

– Size and usage of file system cache

Note: If you see CPUs excessively being in a WAIT state, it means there is

a bottleneck in some of your I/O devices. You should avoid this situation.

Take a closer look at the measurement counters for disks and network.

Chapter 11. Performance monitoring, analysis, and tracing 285

 Operating system kernel statistics:

– Length of run queue: the number of threads waiting to be scheduled to

have some processor assigned.

– Load: the number of processes ready to be performed by some processor;

typically this is referred to as an average over some preceding time interval.

A long run queue typically goes hand in hand with a high load.

 Network statistics

For each of the network interfaces:

– I/O rate in terms of number of data packets and bytes in both directions (in

and out)

– Collisions and errors on this network interface

 Disk I/O statistics

For each of the disks attached to your computer:

– I/O rate in terms of the number of data packets and bytes in both directions

(in and out)

Note: If your paging space is being used up to the limit during regular

operating hours, this means that the system might not have enough paging

space during the peak time. You should avoid this situation.

In addition, if you see a lot of I/O to your paging space, this increases

WAIT-time and consumes capacity of your I/O devices. You should avoid

this, too. You might need to add more physical memory to your hardware.

Note: If you constantly see a long run queue, it means that each thread

has to wait for a longer time before being scheduled for some CPU. You

should avoid this situation.

For the same reason, you should avoid a situation in which you see a

constant high load on your system.

Note: If you see constant data transfer rates on any of your network

interfaces that come close to the interface’s capacity, it means this network

interface is about to be a bottleneck. It is very likely that you observe a high

value for CPU in WAIT state. You should avoid this situation.