You can use the command language to create batc h ￿les (also called stream jobs) that con tain control statemen ts and variables. Execution of the commands in the ￿le can be altered at execution time b y using these control statemen ts.

MPE/iX System Performance

MPE/iX provides e￿cient performance through use of the mapped ￿le tec hnique and concurrent directories described in the follo wing subsections.

Mapped Files

MPE/iX emplo ys the mapped ￿le tec hnique for performing ￿le access. It is an impro ved version of the disc cac hing-capabilit y of MPE V/E. File access e￿ciency is impro ved when code and data portions of ￿les required for processing reside in memory . Accessing memory is faster than performing ph ysical disc I/O operations. The mapped ￿le tec hnique eliminates ￿le system bu￿ering and optimizes global system memory managemen t.

File mapping is based on MPE/iX demand paged virtual memory , which uses to advantage the large amount of virtual memory on the system. When a ￿le is opened, it is logically mapped into virtual memory. An open ￿le and its con tents are referenced by virtual addresses. Eac h byte of each opened ￿le has a unique virtual address.

File mapping impro ves I/O performance without imposing additional CPU o verhead or sacri￿cing data in tegrity and protection. Traditional disc cac hing schemes for increasing I/O performance impose a CPU o verhead penalty. The 900 Series hardw are and system architecture allow MPE/iX to perform ￿le mapping without incurring this penalt y. System hardware performs the virtual to ph ysical address translations for locating portions of the mapped ￿les, thus eliminating CPU o verhead for this function.

If the required pages are not in memory , the MPE/iX Memory Manager fetc hes them directly from disc and places them in the user's area in memory . This eliminates File System bu￿ering. Pages are "prefetched" to reduce the amoun t of physical disc I/O. Prefetc hing means that the page speci￿ed for fetc hing and the group of pages surrounding it are a fetc hed all at once. This impro ves e￿ciency because the processor is likely to require pages that are located near each other. Two bene￿ts of this are:

Eliminating unnecessary data mo vement in memory impro ves system performance.

Memory space usage is optimized.

MPE/iX ￿le system access in trinsics are built on the mapped ￿le tec hnique. Programs using ￿le access methods supported b y MPE ￿le types and intrinsics obtain the bene￿ts of ￿le mapping without requiring c hanges.

You can directly access mapped ￿les when programming in languages with poin ters. For example, you can obtain the advantage of File System naming and data protection for accessing array t ype structures and dev eloping specialized access methods.

You can write programs that address ￿les through virtual memory , instead of calling File System intrinsics for disc reading and writing. The ￿le in terface provides opening and closing of user mapped ￿les with normal naming and securit y, but with impro ved LOAD and STORE speed on ￿le references.

OVERVIEW 1-13