CICS - Programming - Programming Considerations

Programming Considerations

Multiple-user interactive-transaction application programs were required to be quasi-reentrant in order to support multiple concurrent transaction threads. A software coding error in one application could block all users from the system. The modular design of CICS reentrant / reusable control programs meant that, with judicious "pruning," multiple users with multiple applications could be executed on a computer with just 32K of expensive magnetic core physical memory (including the operating system).

Considerable effort was required by CICS application programmers to make their transactions as efficient as possible. A common technique was to limit the size of individual programs to no more than 4,096 bytes, or 4K, so that CICS could easily reuse the memory occupied by any program not currently in use for another program or other application storage needs. When virtual memory was added to versions OS/360 in 1972, the 4K strategy became even more important to reduce paging and thrashing unproductive resource-contention overhead.

The efficiency of compiled high-level COBOL and PL/I language programs left much to be desired. Many CICS application programs continued to be written in assembler language, even after COBOL and PL/I support became available.

With 1960s-and-1970s hardware resources expensive and scarce, a competitive "game" developed among system optimization analysts. When critical path code was identified, a code snippet was passed around from one analyst to another. Each person had to either (a) reduce the number of bytes of code required, or (b) reduce the number of CPU cycles required. Younger analysts learned from what more-experienced mentors did. Eventually, when no one could do (a) or (b), the code was considered optimized, and they moved on to other snippets. Small shops with only one analyst learned CICS optimization very slowly (or not at all).

Because application programs could be shared by many concurrent threads, the use of static variables embedded within a program (or use of operating system memory) was restricted (by convention only).

Unfortunately, many of the "rules" were frequently broken, especially by COBOL programmers who might not understand the internals of their programs or fail to use the necessary restrictive compile time options. This resulted in "non-re-entrant" code that was often unreliable, leading to spurious storage violations and entire CICS system crashes.

The entire partition, or Multiple Virtual Storage (MVS) region, operated with the same memory protection key including the CICS kernel code. Program corruption and CICS control block corruption was a frequent cause of system downtime. A software error in one application program could overwrite the memory (code or data) of one or all currently running application transactions. Locating the offending application code for complex transient timing errors could be a very-difficult operating-system analyst problem.

These serious shortcomings persisted for multiple new releases of CICS over a period of more than 20 years. CICS application transactions were often mission-critical for public utility companies, large banks and other multi-billion-dollar financial institutions. Top-quality CICS skills were in high demand and short supply. The complex learning curve was shallow and long. Unqualified novice developers could have a major negative impact on company operations.

Eventually, it became possible to provide a measure of advance application protection by performing all testing under control of a monitoring program that also served to provide Test and Debug features. One such software offering was known as OLIVER, which prevented application programs corrupting memory by using instruction set simulation of the application code, providing partial virtualization.