Hardware contentions
Contentions can occur on processor cycles, real storage, database associated hardware I/O operations, and network-associated hardware operations.
- Processor cycles. It is not uncommon for transactions to execute more than one million instructions. To execute these instructions, transactions must contend with other tasks and jobs in the system. Sometimes these tasks and jobs must wait for activities such as file I/O. Transactions give up their use of the processor at these points and must contend for use of the processor again when the activity has completed. Dispatching priorities determine which transactions or jobs get use of the processor, and batch or other online systems affect response time by receiving preferential access to the processor. Batch programs that access online databases also tie up those databases for longer periods of time if their dispatching priority is low. At higher usages, the wait time for access to the processor can be significant.
- Real storage (working set). Just as transactions must
contend for the processor, they also must be given a certain amount
of real storage. A real storage shortage can be particularly significant
in CICS® performance because a normal page fault
that occurs when acquiring real storage results in synchronous I/O.
The basic design of CICS is asynchronous, which means
that CICS processes requests from multiple tasks
concurrently to make maximum use of the processor. Most paging I/O
is synchronous and causes the MVS task that CICS is
using to wait, and that part of CICS cannot
do any further processing until the page operation completes. Most,
but not all, of CICS processing uses a single MVS task (called
QUASI
in the dispatcher statistics). - Database-associated hardware (I/O) operations. When data
is being accessed to provide information that is required in a transaction,
an I/O operation passes through the processor, the processor channel,
a disk control unit, the head of string on a string of disks, and
the actual disk device where the data resides. If any of these devices
are overused, the time taken to access the data can increase significantly.
This overuse can be the result of activity on one data set, or on
a combination of active data sets. Error rates also affect the usage
and performance of the device. In shared DASD environments, contention
between processors also affects performance. This, in turn, increases
the time that the transaction ties up real and virtual storage and
other resources.
Large amounts of central and expanded storage, very large data buffers, and keeping programs in storage, can significantly reduce DB I/O contention and somewhat reduce processor utilization while delivering significant internal response time benefits.
- Network-associated hardware operations. The input and output messages of a transaction must pass from the terminal to a control unit, a communications link, a network controller, a processor channel, and finally the processor. Just as overuse of devices to access data can affect response time, so excessive use of network resources can cause performance degradation. Error rates also affect performance. In some cases, the delivery of the output message is a prerequisite to freeing the processor resources that are accessed, and contention can cause these resources to be tied up for longer periods.