Inherent in the design of your online IMS system are your performance objectives. Establishing
performance objectives is a major task requiring data gathering for
the IMS workload as a whole.
After defining the workload and estimating the resources
required, you must reconcile the desired response with the response
you consider to be attainable. Monitor the performance of the system
to determine if these objectives are being met. Base performance objectives
on:
- Desired, acceptable, and maximum response time
- Average and maximum resource demands (or workload) per transaction
- Predicted and actual transaction volumes
Establishing your performance objectives is an iterative
process involving the following activities:
Procedure
- Defining user-oriented performance objectives and priorities.
These derive from the way an end user perceives the service
provided by the system. For IMS these
objectives state expectations of response time as seen by the end
user at the terminal.
When response time objectives are established,
they should not only reflect the time-in-system (the elapsed time
from entry of a last input message segment to the first response segment),
but also the expected amount of IMS and
application program processing. You should consider whether to define
your criteria in terms of the average, the 90th percentile, or even
worst-case response time. Your choice depends on your installation's
audit controls and the nature of the specific transactions.
- Determining how performance against these objectives is
measured and reported to users.
This includes identification
of systematic differences between the measured data and what the user
sees. You should investigate the differences between internal (as
seen by IMS) and external (as
seen by the end user) measures of response time. To do this, you can
use the following tools:
- The Transaction Response report produced by the IMS Statistics Analysis utility, which gives
the following data on internal response time by transaction type:
- Longest and shortest response
- 25th, 50th, 75th, and 90th percentiles of the response time distribution
- Installation-written programs, which can analyze the output of
the IMS Log Analysis utility.
Other programs can also provide the same type of information, and
are usually tailored to satisfy the installation's requirements.
- Understanding and documenting the current workload.
This requires breaking down the total work into categories
and, for each category, developing a workload profile (generally estimates)
that include:
- Definition of the transaction category (for example, transaction
type or group of transactions). The two characteristics of the category
are:
- The IMS workload, which is
generally documented by transaction profile. In a well-designed IMS online system, most transactions
perform a single function and have an identifiable workload profile.
Later in the process, transaction types with common profiles can be
amalgamated for convenience.
- The transaction volume. In situations where the workload to be
documented is already operational, a summary of transaction volumes
can be obtained from the IMS Statistical
Analysis utility. (The Application Accounting Report gives transaction
counts within program name.) In other cases, volumes are estimated.
- Relative priority of category, including periods during which
priority changes.
- Resource requirements of the work:
- Physical resources managed by the operating system (real storage,
DASD I/O, terminal I/O)
- Logical resources managed by the subsystem, such as control blocks,
latches, buffers, and number of regions
To obtain a base profile of transaction resource demands,
you can start IMS on a dedicated
machine and execute a few transactions to accomplish initialization
and buffer pool usage. Then start the IMS Monitor
and measure a sample of transaction execution.
You can use
the base transaction profile to examine the transaction workload to
see if it can be reduced. Such design changes result in the greatest
impact, occurring before system-wide contention. You can also compare
the base profile to the transaction profile in the production environment.
- Translating the resource requirements and volume information
obtained into system-oriented objectives for each work category
This includes statements about the transaction rates to be
supported (including any peak periods) and the internal response time
profiles to be achieved.
- Confirming that the system-oriented objectives are reasonable.
After initializing the system and monitoring its operation,
you need to find out if the objectives are reasonable (given the hardware
available), based upon the measurements of the workload. If the measurements
differ greatly from the estimates used, you must revise the workload
documentation and system-oriented objectives accordingly, or tune
the system.
- Establishing performance objectives.
Establishing
performance objectives is also a necessary prerequisite to using the z/OS® Workload Manager. Much of
the information gathered when you establish performance objectives
can be used as input when planning for workload management.