AIX Parallel System Support Programs (AIX PSSP) is a collection of administrative software applications designed to provide the unique system management functions required to manage the Scalable POWERparallel 2 (SP2) system as a full function parallel system. AIX PSSP provides extensions to AIX and contains all of the software needed to install, operate, and maintain an SP2 system from a single point, the control workstation. Multiple components are included which are integrated together into a single software program product. In addition, a number of useful publicly available software packages are distributed with AIX PSSP.

AIX PSSP components span the SP2 software framework including:

• System Administration: Installation of AIX/6000 and other system software in the SP2 parallel environment, user and password management, job accounting, print management, system startup/shutdown, and file collection support.
• System Monitor: Performs monitoring and control of the SP2 and key hardware variables.
• System Data Repository: Provides repository for administrative data needed to define and manage the SP2 system.
• Resource Manager: Interfaces with other SP2 job management software to allocate SP2 resources (node quantity, node type, disk) for serial and parallel jobs and monitors and controls their execution.
• Communication Subsystem Support: Provides the functions needed to support the SP2 High-Performance Switch including switch fault handling and parallel application communications interfaces.
• Virtual Shared Disk: Allows creation of logical disk volumes accessible from multiple nodes executing tasks of a parallel application.

The POE consists of components for developing, executing, debugging, profiling, and tuning parallel programs. POE eases the transition from serial to parallel processing by allowing the use of standard AIX tools and techniques. This includes the areas of program compilation, scheduling, execution and monitoring in a manner familiar to the UNIX programmer.

Parallel Desktop:

POE has an AIX Windows/6000 Desktop interface, Parallel Desktop, which simplifies use of POE. Source files, executables, and AIX Parallel Environment tools are represented on the Parallel Desktop as icons, easily manipulated to compile and execute parallel programs or start other AIX Parallel Environment tools. The Partition Manager of POE allows a user to compile his program using the XL FORTRAN or C Set ++ (C and C++) compilers and link in all required run-time libraries. It also controls parallel program execution by allocating nodes required to run the job, copies the executable tasks into the individual allocated nodes and invokes their execution, monitors the status of task execution, and communicates with job scheduling systems.

System Analysis Tools:

POE contains two X Window System analysis tools: Program Marker Array and System Status Array.

The Program Marker Array

is a run-time analysis tool allowing the monitoring of a program's execution by presenting a graphical representation of the program tasks. This window representation consists of a number of small squares called lights that change color under program control. Each parallel task has its own row of lights. Subroutine calls inserted into these tasks change the light colors. This visual feedback can be invaluable in detecting application problems.

The System Status Array

is an on-line analysis tool for examining the utilization of processor nodes. A window consisting of a number of colored squares representing SP2 or RISC System/6000 cluster processor nodes shows the percent of CPU utilization so the nodes that are least busy may be better utilized.

For identifying problem areas in parallel programs and to assist in tuning them, the Parallel Operating Environment contains the familiar AIX profilers, prof and gprof, allowing the determination for each executed task of: Execution time in each routine Number of times the routine was called Average milliseconds per call.

Visualization Tool:

The AIX Parallel Environment's Visualization Tool (VT) is designed to graphically show the performance and communications characteristics of an application program, (Trace Visualization), and also to act as an on-line monitor, (Performance Monitoring). Essentially, the VT is a group of displays, or visuals, which each show some unique execution characteristic of an application or system.

Debugger:

The AIX Parallel Environment builds on the familiar and effective AIX dbx debugging facility but adds additional functions specific to parallel program debugging. The Parallel Debugger provides a command line interface, pdbx and a graphical user interface, xpdbx designed for users not familiar with dbx. The Parallel Debugger lets you:

• 'Unhook' program processes not to be monitored
• Debug object files
• Set breakpoints at selected statements or run a program line-by-line
• Debug a program using symbolic values, displaying them in the correct format.

The Parallel Message Passing Library is a programming interface for converting a serial FORTRAN, C, or C++ program into a parallel application via subroutine calls. The MPL provides a rich and diverse set of subroutines for coding simple operations, such as task-to-task message passing, as well as the more advanced operations required for highly complex communications. At execution time, the user specifies whether parallel task communications can occur via the Internet Protocol (IP) through token ring, ethernet or SP2 High-Performance Switch adapters (configured for IP) or via 'User Space' access to the SP2 High-Performance Switch for maximum performance. Collective communication functions of MPL are subroutines for frequently needed operations involving multiple tasks. For example, the 'scatter' call lets you distribute messages from a single source task to each task in a group. Conversely, with the 'gather' call, you can gather messages from each task in a group into a single destination task. These capabilities can greatly reduce the time and effort spent on developing parallel applications.