Design considerations
The length of time between data set reorganizations can affect performance. The efficiency of access decreases as the data set becomes increasingly fragmented. Fragmentation can be kept to the minimum by reducing the length of time between data set reorganizations.
The following factors can limit performance:
- Database design. A data set or database needs to be designed to meet the needs of the application it is supporting. Such factors as the pattern of access to the data set (especially whether it is random or sequential), access methods chosen, and the frequency of access determine the best database design. Such data set characteristics as physical record size, blocking factors, the use of alternate or secondary indexes, the hierarchical or relational structure of database segments, database organization (HDAM, HIDAM, and so on), and pointer arrangements are all factors in database performance.
- Network design. This item can often be a major factor in response time because the network links are much slower than most components of an online system. Processor operations are measured in nanoseconds, line speeds in seconds. Screen design can also have a significant effect on overall response time. A 1200-byte message takes one second to be transmitted on a relatively high-speed 9600 bits-per-second link. If 600 bytes of the message are not needed, half a second of response time is wasted. Besides screen design and size, such factors as how many terminals are on a line, the protocols used (SNA, bisynchronous), and full-duplex or half-duplex capabilities can affect performance.
- Use of specific software interfaces or serial functions. The operating system, terminal access method, database manager, data set access method, and CICS® must all communicate in the processing of a transaction. Only a given level of concurrent processing can occur at any one time, and this can also cause a performance constraint. Examples of concurrent processes include the SNA receive any pool (RAPOOL), VSAM data set access (strings), CICS temporary storage, CICS transient data, and CICS intercommunication sessions. Each of these can have a single or multiserver queueing effect on a transaction's response time, and can tie up other resources by slowing task throughput.
One useful technique for isolating a performance constraint in a CICS system with SNA is to use the IBMTEST command issued from a user's terminal. This terminal must not be in session with CICS, but must be connected to the z/OS® Communications Server for SNA.
IBMTEST (n)(,data)
where n is
the number of times you want the data echoed, and data consists
of any character string. If you enter no data, the alphabet and the
numbers zero through nine are returned to the terminal. This command
is responded to by SNA LU.IBMTEST is an echo test designed to give the user a rough idea of the z/OS Communications Server component of terminal response time. If the response time is fast in a slow-response system, the constraint is not likely to be any component from the z/OS Communications Server onward. If the response time is slow, the z/OS Communications Server or the SNA network may be the reason. This sort of deductive process in general can be useful in isolating constraints.
To avoid going into session with CICS, you may have to remove APPLID= from the LU statement or CONNECT=AUTO from the TERMINAL definition.