Performance Guide

The response time of a given system or application is made up of four basic elements

	Processing time, this the time the CPU is executing instructions on behalf of the application.
	Memory, this is the time the application is waiting for memory, or in a page-wait state.  This includes memory pages being retrieved from disk
	I/O, the time the application is waiting for disk, excluding memory related I/O
	Communications, the time involved in the transfer of the data across the network to the requesting application.

All things being equal the faster the processor the more instructions can be executed per second.  There are many benchmarks and standards for measuring and comparing CPU speeds - most are meaningless in the real world.

At a given time the processor can be one of three states

There are two real areas of concern when it comes to memory.

Paging: Paging in a Unix system is not generally a problem, it only becomes a problem when it is excessive. The speed of the disk sub-system can be seriously degraded when large amount of paging occur.

Swapping: If the memory requirements for the application exceeds the memory that is available then the kernel can be start swapping. Swapping will only start to occur when the memory is very low, any swapping shows that the system is very short of memory.

In most cases most, if not all the following will need to be measure and analysed.

The response rate is probably the easiest to understand and to measure but it is the sum of, potentially, a large number of separate elements, and everybody involved with the system has their own interpretation of the response time. The end user, for example, might consider the response time to be the time taken to complete a series of complex task, whereas the DBA might consider each element of that task to have it's own distinct response time.

The system response time, or internal respose time, is the time the system takes from receiving an instruction to do something to the point where it completes it's task. This is time most DBA and administrators would define as the response time, however, it does not include any communication delays that the user might experience. This can, and often does, lead to performance perception problems between the administrators of the system and the end user community.

Most hardware vendors provide their own set of measuring tools that are specific to their version of the operating system, eg Sun has proctool, IBM has perf. Fortunately, there are a number of standard Unix utilties that exist on most platforms althought the exact measurement that is being taken might vary. The most common and useful utilities are:

It is important to realise that although most Unix vendors offer these tools they don't always report the same data, and in some cases the column title might be historical and bear little resemblance to the parameter being reported.

Informix also provides a number of utilites for monitoring performance